A new soft computational approach for discrimination of odors/gases is presented. The proposed technique is applied on the raw data obtained from the responses of oxygen plasma treated thick film tin oxide sensor array exposed to four different odors/gases. The data generated from the sensor array response were subjected to wavelet transform and appropriate coefficients were selected using multiscale principal component analysis (MSPCA). The training and test performances of backpropagation trained neural network (BPNN) and radial basis function neural network (RBFNN) have been compared. Both the networks have been found to identify the odors/gases with a high success rate.