Weird encoding behavior

Hi, I'm experiencing a weird behavior when storing latin characters to a 
PostgreSQL 8.3.1.876 server. The database is Latin1 encoded, and it is 
working since September 2008, it wasn't updated nor replaced since its 
first installation.

The weirdness of the problem is that sometimes the characters are stored 
correctly, but sometimes doesn't (allways by the same program), the 
field type is Varchar(30), and for example the text "NUÑEZ" is stored as 
"NU?EZ".

The data comes from an external application in an XML file (also 
Latin1), then, a Delphi service parses the XML and create the 
Insert/Update statements to store the data in the database. I'd try to 
reproduce the bug by sending XML files with 'Ñ' to the service, but it 
is stored correctly.

Also, there's a front end that allows users to see/edit the data in a 
user friendlier way. Again, I checked by inserting records with 'Ñ' 
using this front-end, and also are stored correctly.

Does anyone faced the same problem? any workaround?

Thanks in advance,
Leonardo