Wow, that is weird. However, I just fixed some bugs relating to the memory that is allocated when reading compound types, so you should get the latest snapshot and try again: http://www.unidata.ucar.edu/software/netcdf/builds/snapshot/netcdf-4/ I know there is still a bug in there related to compound types and their sizes and layouts in memory, when dealing with machines of different architecture. I am working on that one next... ;-) Thanks, Ed Ticket Details =================== Ticket ID: NON-647286 Department: Support netCDF Priority: Normal Status: Closed