This paper introduces a comparison study of three existing methods for building Recursive Deterministic Perceptron Neural Networks. Three methods were compared in terms of their level of generalisation, convergence time and topology sizes. Prior to this study only an exhaustive, NP-Complete method was used for building RDP neural networks. Due to its high complexity, this limited its use in real world classification problems. This work shows that the other two methods, with a polynomial time complexity, can be used as an alternative. These results will widen the use of the RDP neural network. The impact factor is 2.000.