>Mario, that is indeed very strange. We are doing something similar, i.e. using DBFs as file containers or as a 'filesystem within the filesystem' - each of the DBFs holds on the order of 10^5 image files in memo fields for convenient (and speedy) retrieval via some numeric key.
>
>However, we have never run into problems such as you have described. The only noteworthy difference that I can see is that we pass the data around as memory handles (neatly wrapped into objects that take care of freeing the handle) instead of as strings, and the transfer to and from the memo fields happens via low-level API functions like _MemoChan(), _FRead() etc.
>
>The reason we use re-opening of the tables instead of SYS(1104) is that the latter works rather indiscriminately while re-opening the tables affects nothing but the tables themselves - and a subsystem like an image manager should not affect the operation of other subsystems within an application.
>
>Back to your problem: have you been able to isolate the problem and rig some repro code? From what you have reported this looks very much like a grade-A memory leak.

Stephan,
I'll completely rebuild an application tomorrow with only that part in it istead of adding it the the master PRG and I'll let you know if it change anything. I've try the APPEND MEMO option that Sergey give but still the same thing it's very strange.