I have a file that *should* be all ISO-latin, but the program that created it seems to sprinkle UTF-8 characters round in it.
If that's the case, then that program is horribly broken and I would recommend trying to see what you can do to fix that. Anyway, do you have any sample data that shows both UTF-8 and Latin-1 data?
EF BB BF is the Byte order mark encoded as UTF-8, it can be present at the beginning of UTF-8 encoded files. The fact that in your second sample it appears after a series of commas could mean that the program is trying to write a CSV file and used a UTF-8 encode function that adds the BOM on individual fields, or it slurped a file with the wrong encoding and used that as the contents of the field. If this guess is correct, then maybe a solution would be to first parse the CSV file and then individually decode the fields with different encodings, though I would consider that a pretty ugly workaround, plus you'd have to know the encodings (or guess them, which is a workaround in itself).
Can you tell us more about this program, the file format it is outputting, and give more example data that shows the problem?
Update just to address the question in the title and node: simply clobbering any non-ASCII characters without understanding the input data is almost never the right solution, because you'll almost certainly also delete important characters. Instead, first fix the encoding problems, and if you then really want to ASCIIfy your (correctly decoded!) Unicode data, you can use e.g. Text::Unidecode.