A novel technique is presented for the discrimination of Integrated Circuits (ICs) which involves an analysis of the complex reflection coefficient of RF signals (300 kHz to 500 MHz) on the pins of the device. Identical ICs in 8-pin SOIC packages are placed in the test socket of a custom test setup involving an electric network analyzer (ENA) and a 1x8 relay matrix. A measurement of the reflection coefficient for the device under test (DUT) is used to calculate the root mean square error (RMSE) between the DUT and a measurement taken from the empty socket. This is completed for each of the possible states of the relay matrix resulting in a 1x256 feature vector which can be used in a classification analysis. A multivariate logistic regression is used to generate the classification models. These models are trained using synthetic samples generated from the statistical distribution of samples collected empirically. DUTs were chosen from a set of two device families including several part numbers per family and multiple lot IDs per part number. Using this technique to identify devices resulted in an accuracy of 100% when identifying device family and part number and an accuracy of 77% when discriminating between devices with identical part numbers but different lot IDs.