In this work, wire electro-discharge machining of WC-Co composite with two different weight percentages of Co, and a newly developed TiC reinforced austenitic manganese steel matrix composite with two different volume percentages of TiC has been studied. Parametric analysis has been done to investigate the influence of the process parameters on the performances of the process, namely cutting speed, surface roughness, and kerf width. The eroded surface of the WC-Co composite has been investigated by scanning electron microscope. The SEM micrographs show that the machined surface is characterized by recast layer, micro cracks, micro voids, and loosely bound WC grains. As the discharge energy increases, the size of the micro cracks also increases. The process has been modeled by neural networks (NN), namely normalized radial basis function network (NRBFN) with two advanced clustering techniques. There is no work available on the modeling of the wire-EDM process for machining composite materials by these NN techniques. The process has also been modeled by conventional regression technique, standard back-propagation neural network (BPNN), and NRBFN with traditional k-means clustering techniques, to see the effectiveness of the new techniques. The comparison between the models has been made based on the prediction accuracy, number of iterations required to converge, and the ability to provide the parametric analysis of the process. It has been seen that all the models are capable of predicting the performance measures with permissible error limit. The models are also capable of performing the parametric analysis. Irrespective of the work piece materials, it has been observed that, BPNN model yields the best results. Among three different clustering techniques in NRBFN, clustering by gradient descent technique provides the best result, followed by NRBFN with enhanced k-means clustering technique, NRBFN with traditional k-means clustering technique. NRBFN with traditional k-means clustering technique requires the lowest number of iterations. Therefore, for online monitoring of the performances of the wire-EDM process, this technique can be used. The parametric analysis reveals that, there is no parametric combination exists that can simultaneously result in both the best cutting speed and best surface finish quality or the best cutting speed and best dimensional accuracy. Therefore, the optimization of the wire-EDM process parameters has to be considered as multi-objective optimization problem. Neuro- NSGA-II has been developed to obtain pareto-optimal solutions which can satisfy the requirement of the process engineers. The nondominated solutions have been obtained and reported. Keywords: WC-Co, TiC reinforced austenitic manganese steel matrix composite, wire-EDM, NN, NRBFN, BPNN, clustering, neuro-NSGA-II.