Its simplicity makes it is easy to understand, but limitations include:

• No real data structures, just multidimensional arrays and lists
• No nested structures, variable-length types, or ragged arrays
• Only one shared unlimited dimension for appending new data
• A flat name space for dimensions and variables
• Character arrays rather than strings
• A small set of numeric types

In addition, the classic netCDF format has performance limitations for high performance computing with very large datasets:

• Large variables must be less than 4 GB (per record)
• No real compression supported, just scale/offset packing
• Changing a file schema (the logical structure of the file) may be very inefficient
• Efficient access sometimes requires data to be read in the same order as it was written
• Big-endian bias may hamper performance on little-endian platforms
• I/O is serial in Unidata netCDF-3 (but see Argonne/Northwestern Parallel netCDF project)

Despite these limitations, netCDF-3 is very widely used in climate modeling, ocean science, and atmospheric science, and has been used to represent some very complex data, e.g. grids described in Gridspec specification.

2011 Unidata NetCDF Workshop > The Two NetCDF Data Models