The N-tuple approximation network offers many advantages over conventional neural networks in terms of speed of operation and its ability to realise arbitrary nonlinear mappings. However, its generalisation/selectivity properties depend strongly on the form of input encoding being used in the system. The paper analyses the suitability of use of the CMAC code for the N-tuple networks, and compares its properties with existing schemes. It is argued that the application of this type of encoding can provide desirable monotonic mapping between input and pattern space distances without the penalty of very long binary patterns as is the case for bar-chart encoding. Additionally, similarities between the classic N-tuple and CMAC networks are highlighted.