• Not checking the return from each netCDF function call: a non-zero return means an error occurred, and indicates the cause. A function is available to turn error codes into meaningful error messages.
• When writing data, not explicitly calling a netCDF close function before exit, to make sure all the data gets written to disk
• When reading data, not providing a big enough array into which data will be read: because C and Fortran APIs don't include a parameter for how much space the caller is providing, this cannot be checked by compilers.
• When writing data to a variable that uses the unlimited (record) dimension: you usually need to specify how many records to write.
• When reading data from a record variable, specify how many records to read to avoid overflowing a buffer, in case the file has more records than you expect.
• If reading character data using the C API, make the character array into a string by explicitly adding null termination, even for attribute values.

2012 Unidata NetCDF Workshop > Using netCDF in programs