The implementation of Affective Computing concepts requires the assessment of the affective states in the computer user, e.g., "relaxation" or "stress". Traditionally, the Galvanic Skin Response (GSR) signal has been analyzed as the leading indicator of the sympathetic activation that accompanies "stress", when it is experienced by a computer user. However, recent research has found that the Pupil Diameter (PD), which is also controlled by the Autonomic Nervous System (ANS), can be an important indicator of sympathetic activation. This paper describes techniques for the processing of the Pupil Diameter (PD) signal to detect episodes of mental stress induced in experimental subjects, differentiating them from "relaxation" intervals. Our experiments also recorded and analyzed the GSR signal from the subjects, for comparison purposes. The PD signal is first pre-processed applying wavelet denoising and Kalman filtering to remove the high-frequency variations of the raw PD signal that are not representative of the subject's affective state. Then 3 features are extracted from the normalized, pre-processed PD signal and five different classification algorithms are applied on these features to differentiate the states of "relaxation" vs. "stress" in the computer user. Similarly, 3 GSR features were obtained and used for classification. PD-based classification achieved an average accuracy of 85.86%. GSR-based classification achieved an average accuracy of 60.66%. Therefore, the results indicate that the pupil diameter may be one of the most significant physiological signals to monitor for affective assessment and differentiation of "relaxation" vs. "stress" states of a computer user.