Permeability prediction from well logs is of great importance in reservoir characterization and engineering. In this paper, a new method is proposed to correlate conventional well logs and core permeability data. It uses an improved "windowing" technique to incorporate adjacent core data to the permeability predictor in such a way that the scales of the well log and core measurements are matched It also has the capability to evaluate the reliability of each and every prediction. The method is implemented by the use of a neural network and is demonstrated by means of a case study. The study uses a set of well logs and limited core permeability data to produce continuous permeability profiles. The results show that the permeability profiles are consistent with the core permeability and the geological sequence of the reservoir The reliability indicator is particularly useful for examining reservoir heterogeneity and sampling.