If you look at some of the sample patch files, you might notice that, whereas for patching a small (1 byte, 2-hex-digit) value like $32 you'd use the following:
When it comes to larger (1 word, 2 byte, 4-hex-digit) values, such as $1F9A, the number must be split up into two sections, which are then reversed, like this:
It gets more confusing when you want to patch even larger (1 dword, 2 word, 4 byte, 8-hex-digit) values, like $01B3902C, which would be done like this:
All this is due to the way that Intel (and compatible) microprocessors store multi-byte numbers. They are stored in so-called 'little-endian' format, with the least-significant byte first. The upshot of this, is that when you are patching a value, you must patch each byte individually, in the reverse of the order which they come in the value.