You have a screenshot, a photograph of a sign, or a PDF a client sent, and you need to know what it is set in — or, more often, you need something free that looks like it. Automatic identification works when the image is clean and fails exactly when the job is real.
Try the automatic route first, and know its limits
Image-matching services compare letter shapes against a database. They need horizontal text, high contrast, clean edges and, ideally, a full alphabet. They do badly with photographs at an angle, script faces where letters join, tight tracking, and anything where the letters have been modified — which is most logos.
If the file is a PDF rather than an image, do not use identification at all. Open the document properties: PDFs record the name of every font embedded in them, which is an exact answer rather than a guess.
The five letters that give a typeface away
When the tools fail, this is what a designer looks at, and it is quicker than it sounds.
- The lowercase g. One storey or two? A double-storey g with a closed lower loop is old-style; a single-storey g is geometric or humanist. It narrows the field faster than any other letter.
- The lowercase a. Double-storey with a curved bowl, or a single circle with a stem? A single-storey a means geometric almost every time.
- The capital R. Straight leg, curved leg, or a leg that kicks out? This is what separates faces that otherwise look identical — it is how Arial gives itself away against Helvetica.
- Terminals on the c, s and e. Cut horizontally, or at an angle? Horizontal reads as a grotesque; angled reads as humanist, however tightly you set it.
- The numerals. Do they all sit on the baseline at capital height, or do some hang below? Old-style figures are a strong signal and appear in a minority of faces.
Then decide what you actually need
Half the time the identification is a means to an end: the real question is what you can legally use that looks like it. If the answer turns out to be a licensed classic, the alternatives section lists open faces that stand in for the commonly specified ones — matched on the shapes above rather than on the name, with a page each explaining which shapes the substitution turns on.
And if you have narrowed it to a category rather than a face, the archive filters by style — geometric, humanist, grotesque, didone, slab and the rest — which is usually how far identification gets you on a photograph anyway.
A shortcut for websites
If the text is on a web page rather than in an image, you do not need to identify anything. Right-click, Inspect, and look at the computed font-family for that element. The browser will tell you the exact name it is rendering, and the network tab will show you the file it downloaded.
Reading the shape instead of guessing the name
Identification tools match outlines. A person matches categories, and the category narrows the field faster than any upload. Four things decide where a face sits, and all four can be read off a single word at a large enough size.
- Stress: put a finger on the thinnest part of the o. Straight up and down is transitional or modern; tilted to the left is old-style.
- Contrast: how different the thickest and thinnest strokes are. Almost none reads as geometric or grotesque; extreme reads as a didone.
- Terminals: how a stroke ends. Cut flat is a grotesque; cut on an angle is humanist; a teardrop is old-style.
- Aperture: how open the c and the s are. Closed apertures are the giveaway of the twentieth-century grotesques, Helvetica included.
Those four answers are usually enough to name the style, and the style pages list every face in this archive under the same vocabulary with a one-line definition of each term. Going from "I know it is a humanist sans" to a shortlist of five is then a filter rather than a search.
Identifying type from a photograph
Most failed identifications are failed photographs. The matching tools want black letters on a white ground, upright, separated, and large. A sign photographed at an angle, a logo on a gradient, or white type on a photograph will defeat every one of them.
- Crop to a few letters rather than the whole word. Two or three distinctive ones beat a full sentence.
- Straighten it. Even five degrees of rotation lowers the match rate noticeably.
- Raise the contrast until the letters are solid black on solid white, and erase anything touching them.
- If letters overlap or connect, separate them by hand — the tools segment the image before they match it.
When the answer turns out to be a paid font
Often the identification succeeds and the licence costs more than the project. That is the ordinary case rather than a failure, and the substitution is a solved problem: the question is which free face matches the thing that made the original right — its proportions, its contrast, its width — rather than which one looks vaguely similar in a specimen. Our alternatives pages name a commercial face, say what a substitute has to get right, and list the ones in this archive that do.
Then prove it before committing: set the actual words of the actual design in the original and the candidate at the same size in the compare tool. A pairing that survives your own headline is a substitution; one that only survives a pangram is a guess.
Write down what you found
The identification is the easy half. Six months later somebody asks what the heading font was, and the answer is in a browser tab that closed long ago.
Keep four things with the project: the family name, where the file came from, the licence, and the exact weights used. Four lines in a text file beside the artwork. For an open font, keep the file itself and its licence text as well — that is permitted and it makes a rebuild in three years a five-minute job rather than another identification exercise.