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. the grids described in Balaji's GridSpec:

2008 Unidata NetCDF Workshop for Developers and Data Providers > The "Classic" NetCDF Data Model