I have been trying to develop CAPS markers for two sets of samples. The SNP was confirmed by repeated sequencing of the cloned framents from the two sets of samples. However, the unique restriction enzyme cuts both the sets of samples.
There could be several issues. It would help to rule out the obvious: double check your restriction enzyme recognition site and make absolutely sure it doesn't recognise both sequences. Then double check your samples and sequence them directly from the PCR. Sequencing from cloned fragments, especially if they have been generated by PCR, can sometimes give you erroneous sequence, because you are sequencing the cloned fragment of a PCR not the majority fragment of the amplification and any errors in that one sequence will be amplified in the cloned fragment. Depending on how you have been analysing your sequence, it is also possible that you might be looking at the cut site rather than the recognition site of the enzyme (embarrassingly, I have done this before). Also double check that the enzyme doesn't have extra activities (star activity) that can be seen with overdigestion. Some of them do. Without looking at the experimental setup and sequence, this is the only advice I can give.
There are other issues like heterozygosity of samples etc that can only be answered by yourself.
I also agree with the explanations provided by 'David' above. Theye are really valid. If you are going for CAPS analysis using PCR products, do the sequencing of the PCR product ( and not the clone)...since some of the clones may have accumulated variation during the PCR thus creating the site. The possibility of stark activity should also be addressed properly. SOme restricition enzymes have sites which can accomodate more than one baes at certain positions, check if you are working with this enzyme...One important thing would we what is your region of interest, is it sinleg copy or multicopy in the genome. In the first case the total PCR product should give clean profile, however in the second case the total PCR product (if amplified from multiple copies in the genome) may give both the profiles (as some copies may have the site and others may not).THis can be either proved by directly analysing the toltal PCR product or analysis of multiple clones....analyze it carefully and you'll have the answer