The input data in the test phase of the odor classifier in which system parameters are trained and fixed in the training phase are usually observed to be drifted due to sensor's aging and/or contamination so the trained Radial Basis Function Network (RBFN) weights can no longer be effective in odor classification. In this paper, enhanced signal processing techniques are introduced for readjusting the RBFN weights in the test phase. The strategy for readjustment is that the test phase output distribution is to follow and match the probability distribution function (PDF) of the target values that were used in the training phase. Instead of using the trained and stored output data as Parzen window samples for the construction of the desired target PDF, which was our initial method, we propose two new methods for the desired PDF construction. One method is to use a set of randomly generated data that the PDF of the generated data matches that of the target values perfectly, and the second one is to use a set of Dirac-delta functions for the target PDF construction. From the experimental results, the proposed methods significantly outperform the previous method in sensor-drift compensation. Copyright © 2011 American Scientific Publishers.