Estimating the petrophysic parameters of reservoir such as porosity, permeability and water saturation for evaluation of hydrocarbon reserve is very important. Despite the developments and the variety of current approaches for determining permeability is steel the most accurate way to experiment on core, which is so time and money consuming. Among these approaches well logging is able to estimate the permeability properly. However this approach encounters lots of problems in some reservoirs. Because of the existence of well logging data for most of wells it is better to be able to estimate permeability with an adequate accuracy via a cheaper approach using well logging data. The aim of this research is determining permeability using well logging data with neural network and wavelet network approaches. Preprocessing and multivariate statistical studies between input and output data was first done in order to eliminate outlier data and to recognize data patterns. Then using well logging data and estimating parameters as input and core permeability values as target ANN and WNN were created and trained to estimate permeability. The results showed that ANN with one hidden layer is more efficient than ANN with more hidden layers. The best network to estimate permeability was made of 11 inputs and the overall fitting between the predicted permeability by network with 27 neurons in one hidden layer and measured core data was 92% in the training phase and 75% in the test phase. The best WNN to estimate permeability was made of 11 inputs and the overall fitting between the predicted permeability by network with 28 neurons in hidden layer and measured core data was 99% in the training phase and 94% in the test phase which is very acceptable. Comparison between the results of these two approaches showed that W because of the ability to model local variations with different scales are more efficient in estimating permeability for regions with severe and local variations.