Research · Choosing a font
Choosing a font: criteria for the right font
This research sets out the criteria for choosing the font of a design system for the Romanian state. It describes the situation on 22 October 2026. Here you will find:
- what studies and standards say about legibility;
- which fonts other states use;
- which languages the font has to cover in Romania;
- how open-licence fonts perform against each criterion, with specimens drawn directly from their files.
0. Summary in eight points
- There is no single "most legible" font. Studies show which features help:
- lowercase letters that are tall enough;
- open shapes;
- letters that are wide enough, with space between them;
- characters that are easy to tell apart.
- The characters most easily confused are capital I, lowercase l and the digit 1, then capital O, the digit 0 and lowercase o. The German legibility standard DIN 1450 and the US government's accessibility guide both require them to be distinct.
- "Dyslexia fonts" do not help. Wider spacing does, and users must be able to increase it.
- In Romania, the font has to write many languages correctly. The state applies the European Charter for Regional or Minority Languages to 20 languages, written in the Latin, Cyrillic, Greek, Armenian and Hebrew alphabets. This rules out most of the fonts presented as "highly legible", including Sora, the font used by ROeID, Romania's electronic identity service.
- A font of your own has a price. States that made their own font cannot pass it on to suppliers and partners, who end up using Arial. Newer systems choose open-licence fonts.
- The font is hosted on the state's own servers, not loaded from Google.
- Of the fonts checked, Source Sans 3 and IBM Plex Sans meet the criteria best. The final choice should be made together with the General Secretariat of the Government (SGG), after a short test with people.
- Section 5 gathers the criteria into a list that can be used for any font choice.
1. What studies and standards say
1.1 Characters that are easy to confuse
The German standard DIN 1450 asks for easily confused characters to be avoided and names I, l, 1 and O, 0. The US government's accessibility guide (digital.gov) says: "I, l, and 1 are distinct. 0 and O are distinct." A review of studies on confused letters adds other pairs: rn and m, cl and d, B and 8, 5 and S. In one of the studies cited there, a digit 1 drawn with a flag and a base was recognised correctly 97% of the time, against 43% without the base.
The I/l/1 and O/0 pairs are the ones standards and guidance agree on. Studies carried out directly on interfaces are, however, few.
Roboto the Android phone font I and l look the same
Inter default I and l look the same
IBM Plex Sans default I, l and 1 differ
Source Sans 3 with serifed I and slashed 0 I, l and 1 differ
The distinction matters most in codes: personal numerical codes (CNP), IBANs, company tax codes (CUI), case numbers, vehicle registration numbers, confirmation codes. There, readers cannot guess from context. There is also a case specific to Romania: the Ukrainian letter І looks identical to the Latin I. A serifed I therefore helps in Cyrillic text too.
Roboto default
Source Sans 3 with serifed I and slashed 0
1.2 Tall lowercase letters
At the same font size, some fonts look 20–30% bigger than others. The reason is the height of the lowercase letters (in typography, the "x-height"). An analysis of 1,049 fonts found it ranges from 0.28 to 0.58 of the font size. The font size alone therefore does not say how big the text looks. A larger x-height helps letter recognition.
Source Sans 3
IBM Plex Sans
Inter
1.3 Open shapes, wide letters, space between letters
- Open shapes. Fonts with open shapes (a, c, e, s) are read faster at a glance than fonts with closed shapes. This is what an MIT study of reading on in-car displays and a comparison of eight sans-serif fonts show. Guidance based on DIN 1450 rates Arial and Helvetica "less suitable" because their shapes are closed.
- Wide letters and space between them. They help people with low vision: in a study of 20 people with macular degeneration, with an average age of 84, reading acuity improved.
- Bold letters help only at very small sizes. Very thin and very bold letters both make recognition harder.
Roboto more closed shapes
Source Sans 3 open shapes
1.4 Dyslexia
Studies of the OpenDyslexic and Dyslexie fonts found no gain in speed or accuracy. With Dyslexie, the small advantage disappeared when the spacing was matched: the spacing helped, not the letter shapes. Wider spacing also helped children with dyslexia in other studies.
1.5 Atkinson Hyperlegible and Luciole
Atkinson Hyperlegible was made by the Braille Institute for people with low vision. It tells letters apart very well, but there is no published study of reading speed or accuracy with it.
Luciole was tested with 145 readers, 73 of them with low vision. About half of those with low vision preferred it, but the measured gain was small. Its text fonts are licensed under CC BY 4.0, which is not a font licence.
1.6 Everyone reads differently
A study of 352 adults and 16 fonts found that each person read 35% faster in their fastest font than in their slowest. The fastest font differed from person to person. 73% of participants wrongly thought their preferred font was also their fastest. Legibility researchers therefore recommend font features, not a particular font, and users should be able to choose their font.
1.7 What accessibility rules require
WCAG, the international web accessibility guidelines, does not require any particular font or size. The European standard EN 301 549 follows WCAG.
Success criterion 1.4.12, however, requires pages to stay usable when users increase:
- the space between lines to 1.5 times the font size;
- the space between letters to 0.12;
- the space between words to 0.16;
- the space between paragraphs to twice the font size.
This concerns the font, but mostly the components: buttons with a fixed height are usually the first to break.
1.8 What is proven and what is not
| Claim | How solid it is |
|---|---|
| Taller lowercase letters and a suitable size help recognition | solid |
| Wide letters and wider spacing help people with low vision and with dyslexia | solid, moderate effect, studies with small groups |
| Open shapes are read faster at a glance | fairly solid |
| I, l, 1 and O, 0 must be distinct | consensus of standards, few direct studies on interfaces |
| Serif or sans-serif fonts read better on screen | no consistent evidence |
| Special dyslexia fonts help | not supported |
| Atkinson Hyperlegible is measurably more legible | no published evidence |
| One font is best for everyone | contradicted |
2. What other states use
| State or system | Font | Licence |
|---|---|---|
| European Commission (ECL component library) | Inter, which replaced Arial | open |
| EU institutions (europa.eu) | Arial | system font |
| USA (USWDS) | Source Sans (default); Public Sans, optional, no longer maintained | open |
| Norway | Source Sans 3 (NAV); Inter, with the tailed l recommended (Designsystemet) | open |
| Finland (Suomi.fi) | Source Sans 3 | open |
| Denmark | IBM Plex Sans, across government | open |
| Germany (KERN, the federal and state standard) | Fira Sans | open |
| Canada | Noto Sans for text, Lato for headings | open |
| Italy | Titillium Web; "Titillium Pro" is planned, to fix legibility problems | open |
| Republic of Moldova (the design system's code) | Onest | open |
| Estonia (TEDI) | Roboto | open |
| United Kingdom (GOV.UK) | GDS Transport, kept after the 2025 brand refresh | gov.uk domains only |
| France | Marianne | reserved for state websites |
| Netherlands; German federal government; NHS | Rijksoverheid Sans; BundesSans; Frutiger | proprietary or paid |
| Romania: the SGG visual identity guide | Trajan Pro (institution names), Trebuchet MS (everything else) | commercial |
| Romania: ROeID | Sora | open, but without Cyrillic letters |
No state uses Atkinson Hyperlegible as its main font. You can find more about European systems in Design systems in Europe.
3. What the font has to cover in Romania
3.1 The languages
| Languages | Characters | Note |
|---|---|---|
| Romanian | ă â î ș ț Ș Ț, with a comma below | not ş ţ with a cedilla |
| Hungarian | ő ű Ő Ű | |
| German, Czech, Slovak, Polish, Croatian, Serbian in Latin script, Albanian, Italian | ß ě ř ů ľ ą ę ł č ć đ š ž ë ç | |
| Turkish and Tatar | ğ ı İ and ş with a cedilla | here, the cedilla is correct |
| Romani | Latin letters, plus θ and ʒ in the spelling used in schools | ʒ is missing from some fonts |
| Ukrainian, Russian, Rusyn, Bulgarian, Serbian, Macedonian | ґ є і ї; the local forms of Bulgarian and Serbian letters | in Bulgarian and Serbian, a few letters (б, г, д, п, т) are drawn differently |
| Greek | the Greek alphabet | |
| Armenian, Yiddish (in the Hebrew alphabet) | separate alphabets | almost never in the same font; a companion font is added |
Source Sans 3 with a comma: correct in Romanian
Source Sans 3 with a cedilla: wrong in Romanian, correct in Turkish and Tatar
Source Sans 3 every letter
Atkinson Hyperlegible Next no Cyrillic or Greek
3.2 The candidates, checked
I downloaded 22 open-licence font families from the official source, Google Fonts, together with their licences. I added Trebuchet MS, Verdana and Arial for comparison. Then I used a program to check, character by character, what each font contains. The data in the table comes from the font files, not from descriptions.
| Font | ș ț, ő ű | Cyrillic | Bulgarian / Serbian forms | Greek | Armenian / Hebrew | I and l, without settings | Slashed 0 | Lowercase height | Equal-width digits |
|---|---|---|---|---|---|---|---|---|---|
| Source Sans 3 | yes | yes | yes / yes | yes | no / no | distinct (tailed l); with one setting, a serifed I too | option | 0.478 (smallest) | yes |
| IBM Plex Sans | yes | yes | yes / no | yes | no / companion font | distinct (serifed I, tailed l) | option | 0.516 | yes |
| Inter | yes | yes | no / no | yes | no / no | the same; distinct with one setting | option | 0.546 | with one setting |
| Noto Sans | yes | yes | no / yes | yes | companion fonts | similar; no tailed l | option | 0.536 | yes |
| Fira Sans | yes | yes | unconfirmed | yes | no | distinct (tailed l) | option | 0.527 | with one setting |
| Open Sans | yes | yes | no / yes | yes | no / yes | similar | option | 0.535 | yes |
| Roboto (the Android phone font) | yes | yes | not checked | yes | no | the same | no | 0.528 | yes |
| Atkinson Hyperlegible Next | yes | no | — | no | no | distinct | default | 0.496 | yes |
| Sora (ROeID) | yes | no | — | no | no | similar | no | 0.534 | yes |
| Lexend, Public Sans, Red Hat Text, Figtree, Titillium Web | yes | no | — | no | no | varies | varies | — | varies |
| Trebuchet MS (SGG guide), commercial | yes | yes | — | yes | no | distinct (tailed l) | no | — | — |
| Arial, commercial | yes | yes | — | yes | no / yes | almost the same | no | 0.519 | — |
The lowercase height is given as a fraction of the font size. All the candidates have ș and ț with a comma and ş and ţ with a cedilla, as separate letters.
What the table shows:
- Source Sans 3 has the best language coverage, including the Bulgarian and Serbian forms and the letters θ and ʒ. Its lowercase letters are, however, the shortest. At 18 px, its text looks about as big as Inter or Verdana at 16 px.
- IBM Plex Sans is the only one that clearly tells I from l without any setting. This matters in programs where you cannot switch on letter variants. It has no Serbian forms and no ʒ, and its boldest weight is lighter than the others' (700).
- Inter has identical I and l without settings. Every application, from the website to Figma and Office, has to be configured. It has no Bulgarian or Serbian forms.
- Noto Sans is very good as a fallback font. It has companion fonts for Armenian and Hebrew.
- Atkinson Hyperlegible Next has no Cyrillic or Greek letters. At most, it could be offered as a "highly legible text" option on pages in Latin script.
4. Practical considerations
- Hosting. The font is kept on the state's own servers, not loaded from Google Fonts. In 2022, a court in Munich awarded 100 euros in damages: a website loaded its font from Google and so sent the visitor's IP address without consent. The court noted that the font could have been hosted on the site's own server. It is a single first-instance decision, but a clear signal. The old argument that the font is "already in the browser's memory" has not held since 2020 either: since Chrome 86, each site has its own separate memory.
- Reserved names. Source Sans 3 ("Source") and IBM Plex Sans ("Plex") have an open licence (OFL) with a reserved name. A modified version, for example one that keeps only the letters you need, has to be renamed. The original files can be used without restriction. Inter has no reserved name.
- Ș and ț in content. Publishing systems should automatically turn ş and ţ with a cedilla into ș and ț with a comma, but only in text marked as Romanian, because the cedilla is correct in Turkish and Tatar.
- The language of each text is marked in the code (
lang="hu",lang="uk",lang="bg"). Only then do the correct forms of Bulgarian and Serbian letters appear and hyphenation work. - Fallback fonts have to be checked too. A fallback font without ș and ț with a comma mixes two fonts in the same word.
- The phone's or computer's own font is not a solution. You do not control the letter shapes, and language coverage differs from one system to another. On Android, for example, I and l look the same. It is good only as a fallback.
- Office documents. I have not checked whether Word and PowerPoint can switch on the letter variants. If the font also has to replace Trebuchet MS in institutions' documents, the default letter shapes matter.
5. The recommendation: criteria for the right font
A font suited to the Romanian state's services should meet all the criteria below.
| # | Criterion | How to check |
|---|---|---|
| 1 | Easily confused characters are distinct: I, l and 1; O, 0 and o | type "Il1 O0o" and a code (a personal number, an IBAN) and compare them at normal text size |
| 2 | Writes every language required in Romania correctly: ș and ț with a comma, ő and ű, Cyrillic with the Bulgarian and Serbian forms, Greek; Armenian and Hebrew through a companion font | type a test line in each language and look for empty rectangles or letters from another font |
| 3 | Tall enough lowercase letters, with body text sized to match | compare fonts at the same size; for body text, at least 16–19 px, depending on the font |
| 4 | Open shapes, letters wide enough, no very thin weights for text | compare a, c, e, s at small sizes |
| 5 | Withstands wider spacing set by the user (WCAG 1.4.12) | apply the wider spacing to real pages and components |
| 6 | Equal-width digits, for tables, amounts and codes | line up a column of amounts |
| 7 | Open licence, with the right to host it, use it in apps and documents and pass it on to suppliers | read the licence (OFL) and check whether it has a reserved name |
| 8 | Maintained and used by other administrations | check the latest versions and who else uses it |
| 9 | Works everywhere: website, phone apps, documents, PDF | test in Word, on Windows with an ordinary monitor, on iOS and on Android |
| 10 | Tested with people: older people, people with low vision, Hungarian and Ukrainian speakers | a short test, on a form with a personal number and an IBAN |
The fonts that meet the criteria best:
- Source Sans 3, for its language coverage. It should be used with two of the font's options: a serifed I everywhere and a slashed 0 in fields for codes. Body text should be at least 18 px. The US, Norway and Finland use it.
- IBM Plex Sans, for letters that are distinct without any setting. It is the safer choice if the font also has to work in institutions' documents. Denmark uses it.
In both cases, Noto Sans Armenian and Noto Sans Hebrew would cover Armenian and Hebrew. Inter remains a good option if matching the European Commission matters a lot. It would then need a version with distinct I and l from the start, which its licence allows.
My proposal, the Government of Romania Design System (SDGR), requires "open fonts with correct ș and ț" and with the characters of minority languages. I would not choose the font alone: the choice should be made together with the SGG, which is responsible for the Government's visual identity, and checked with people.
6. What I could not confirm
- The exact values in the DIN 1450 standard. The standard is paid for, so I used descriptions of it.
- Whether Word and PowerPoint can switch on the letter variants (serifed I, slashed 0).
- How the finalists look on Windows, on an ordinary monitor (96 dpi), at 14–16 px.
- Whether ROeID still uses Sora. The source is about three years old.
- The Bulgarian and Serbian forms of Fira Sans, in the Google Fonts file.
Sources
- DIN 1450, described by leserlich.info: https://leserlich.info/zeichen/schriftart
- Linotype Neue Frutiger 1450, a font made to DIN 1450: https://www.myfonts.com/pages/linotype-neue-frutiger-1450
- digital.gov, the accessibility guide for visual design: https://digital.gov/guides/accessibility-for-teams/visual-design
- Bohm, a review of studies on misrecognised letters and symbols: https://typography.guru/journal/letters-symbols-misrecognition/
- Legge and Bigelow 2011, letter size and fluent reading: https://pmc.ncbi.nlm.nih.gov/articles/PMC3428264
- Beier, Oderkerk, Bay and Larsen 2021, reading in people with macular degeneration: https://thereadabilityconsortium.org/wp-content/uploads/2023/08/IDJ_2021Beier_etal.pdf
- Wallace et al. 2022, different fonts for different people: https://research.adobe.com/publication/towards-individuated-reading-experiences-different-fonts-increase-reading-speed-for-different-individuals
- Reimer et al. 2014 (MIT AgeLab), humanist fonts and reading at a glance: https://dspace.mit.edu/handle/1721.1/96509
- Sawyer et al. 2020, a comparison of eight sans-serif fonts: https://readabilitymatters.org/articles/the-great-typography-bake-off
- Zorzi et al. 2012, spacing and dyslexia: https://pmc.ncbi.nlm.nih.gov/articles/PMC3396504
- Marinus et al. 2016, the Dyslexie font: https://researchers.mq.edu.au/en/publications/a-special-font-for-people-with-dyslexia-does-it-work-and-if-so-wh/
- Galiano et al. 2023, Luciole: https://hal.univ-lyon2.fr/hal-04086507
- Braille Institute, Atkinson Hyperlegible: https://www.brailleinstitute.org/freefont/
- WCAG 2.2, success criterion 1.4.12 (text spacing): https://www.w3.org/WAI/WCAG22/Understanding/text-spacing.html
- Council of Europe, minority languages in Romania: https://www.coe.int/en/web/european-charter-regional-or-minority-languages/-/minority-languages-in-romania-strong-support-in-education-but-the-population-threshold-for-the
- European Commission, ECL component library, typography: https://ec.europa.eu/component-library/ec/guidelines/typography/
- USWDS, typography: https://designsystem.digital.gov/components/typography/
- Denmark, Det Fælles Designsystem: https://designsystem.dk/styleguide/typografi/fonte/
- Norway, Designsystemet: https://designsystemet.no/en/fundamentals/theme/typography; NAV Aksel: https://aksel.nav.no/grunnleggende/styling/typografi
- Finland, the Suomi.fi components: https://github.com/vrk-kpa/suomifi-ui-components
- GOV.UK, the typeface: https://design-system.service.gov.uk/styles/typeface/
- SGG, the visual identity guide (2021): https://sgg.gov.ro/1/wp-content/uploads/2021/09/ghid-identitate-vizuala_SIPOCA35_v2.pdf
- ROeID, the integration guide: https://github.com/roeid-ro/integrare
- The history of ș and ț on computers, by Michael Kaplan: https://archives.miloush.net/michkap/archive/2011/08/24/10199324.html
- The Munich court, Google Fonts and the GDPR: https://www.ra-plutte.de/lg-muenchen-dynamische-einbindung-google-web-fonts-ist-dsgvo/
- Chrome, separate memory for each site: https://developer.chrome.com/blog/http-cache-partitioning
- The OFL licence, frequently asked questions: https://openfontlicense.org/ofl-faq/
- The fonts, with their licences: https://github.com/google/fonts