How can I determine if a file is a PDF file?

Here is what I use into my NUnit tests, that must validate against multiple versions of PDF generated using Crystal Reports:

public static void CheckIsPDF(byte[] data)
    {
        Assert.IsNotNull(data);
        Assert.Greater(data.Length,4);

        // header 
        Assert.AreEqual(data[0],0x25); // %
        Assert.AreEqual(data[1],0x50); // P
        Assert.AreEqual(data[2],0x44); // D
        Assert.AreEqual(data[3],0x46); // F
        Assert.AreEqual(data[4],0x2D); // -

        if(data[5]==0x31 && data[6]==0x2E && data[7]==0x33) // version is 1.3 ?
        {                  
            // file terminator
            Assert.AreEqual(data[data.Length-7],0x25); // %
            Assert.AreEqual(data[data.Length-6],0x25); // %
            Assert.AreEqual(data[data.Length-5],0x45); // E
            Assert.AreEqual(data[data.Length-4],0x4F); // O
            Assert.AreEqual(data[data.Length-3],0x46); // F
            Assert.AreEqual(data[data.Length-2],0x20); // SPACE
            Assert.AreEqual(data[data.Length-1],0x0A); // EOL
            return;
        }

        if(data[5]==0x31 && data[6]==0x2E && data[7]==0x34) // version is 1.4 ?
        {
            // file terminator
            Assert.AreEqual(data[data.Length-6],0x25); // %
            Assert.AreEqual(data[data.Length-5],0x25); // %
            Assert.AreEqual(data[data.Length-4],0x45); // E
            Assert.AreEqual(data[data.Length-3],0x4F); // O
            Assert.AreEqual(data[data.Length-2],0x46); // F
            Assert.AreEqual(data[data.Length-1],0x0A); // EOL
            return;
        }

        Assert.Fail("Unsupported file format");
    }

you can find out the mime type of a file (or byte array), so you dont dumbly rely on the extension. I do it with aperture's MimeExtractor (http://aperture.sourceforge.net/) or I saw some days ago a library just for that (http://sourceforge.net/projects/mime-util)

I use aperture to extract text from a variety of files, not only pdf, but have to tweak thinks for pdfs for example (aperture uses pdfbox, but i added another library as fallback when pdfbox fails)

Pdf files begin "%PDF" (open one in TextPad or similar and take a look)

Any reason you can't just read the file with a StringReader and check for this?

Here is a method that checks for the presence of %%EOF with optional checks for white-space characters. You can pass in either a File or a byte[] object. There is less restriction for white-space characters in some PDF versions.

public boolean isPdf(byte[] data) {
    if (data == null || data.length < 5) return false;
    // %PDF-
    if (data[0] == 0x25 && data[1] == 0x50 && data[2] == 0x44 && data[3] == 0x46 && data[4] == 0x2D) {
        int offset = data.length - 8, count = 0; // check last 8 bytes for %%EOF with optional white-space
        boolean hasSpace = false, hasCr = false, hasLf = false;
        while (offset < data.length) {
            if (count == 0 && data[offset] == 0x25) count++; // %
            if (count == 1 && data[offset] == 0x25) count++; // %
            if (count == 2 && data[offset] == 0x45) count++; // E
            if (count == 3 && data[offset] == 0x4F) count++; // O
            if (count == 4 && data[offset] == 0x46) count++; // F
            // Optional flags for meta info
            if (count == 5 && data[offset] == 0x20) hasSpace = true; // SPACE
            if (count == 5 && data[offset] == 0x0D) hasCr    = true; // CR
            if (count == 5 && data[offset] == 0x0A) hasLf    = true; // LF / EOL
            offset++;
        }

        if (count == 5) {
            String version = data.length > 13 ? String.format("%s%s%s", (char) data[5], (char) data[6], (char) data[7]) : "?";
            System.out.printf("Version : %s | Space : %b | CR : %b | LF : %b%n", version, hasSpace, hasCr, hasLf);
            return true;
        }
    }

    return false;
}

public boolean isPdf(File file) throws IOException {
    return isPdf(file, false);
}

// With version: 16 bytes, without version: 13 bytes.
public boolean isPdf(File file, boolean includeVersion) throws IOException {
    if (file == null) return false;
    int offsetStart = includeVersion ? 8 : 5, offsetEnd = 8;
    byte[] bytes = new byte[offsetStart + offsetEnd];
    InputStream is = new FileInputStream(file);
    try {
        is.read(bytes, 0, offsetStart); // %PDF-
        is.skip(file.length() - bytes.length); // Skip bytes
        is.read(bytes, offsetStart, offsetEnd); // %%EOF,SP?,CR?,LF?
    } finally {
        is.close();
    }
    return isPdf(bytes);
}

Relying on magic numbers does not really appeal to me. I ended up using a preflight library from Apache for this:

compile group: 'org.apache.pdfbox', name: 'preflight', version: '2.0.19'

private boolean isPdf(InputStream fileInputStream) {
    try {
        PreflightParser preflightParser = new PreflightParser(new ByteArrayDataSource(fileInputStream));
        preflightParser.parse();
        return true;
    } catch (Exception e) {
        return false;
    }
}

PreflightParser has constructors for files and other data sources.

In general, we can like this, any pdf version going to finish with %%EOF so we can check like bellow.

public static boolean is_pdf(byte[] data) {
        String s = new String(data);
        String d = s.substring(data.length - 7, data.length - 1);
        if (data != null && data.length > 4 &&
                data[0] == 0x25 && // %
                data[1] == 0x50 && // P
                data[2] == 0x44 && // D
                data[3] == 0x46 && // F
                data[4] == 0x2D) { // -

              if(d.contains("%%EOF")){
                 return true; 
              }         
        }
        return false;
    }