It does sound an awful lot like a wrapping problem. However, with reference to your parenthetical: The bank already makes it impossible to figure out exactly what the original was, by de-wrapping. The reason is that the 27-character lines can be mostly spaces. For example, the dewrapping I gave for the T-DSL example above was how my algo de-wrapped it. The original looks like (pipes added to show end-of-field):

R12345678-A1234567 INCL.EUR|
3,31 MWST JULI             |
MONATL. GEB HR T-DSL FLAT  |
01.07.04-31.07.04     23,99|
[download]

Unfornatly, there is no way of telling that without domain knowladge. (So how do I know that? I load the more-info page on that transaction, but I can't figure out how to do that programatacilly reliably, as the bank keeps changing session IDs. Worse, if I get it wrong, I don't get an error reliably -- I simply get information on the wrong transaction!)

As to using Text::Wrap... well, I was going to point out the bit in the docs where it says "it will destory any whitespace in the original text", but it turns out that only applies to Text::Wrap::fill, not Text::Wrap::wrap. OTOH, I don't see how it could be better then the current solution, and the current solution is already working, tested, and readable.