This paper details the design, development and evaluation of an affective tutoring system (ATS)--an e-learning system that detects and responds to the emotional states of the learner. Research into the development of ATS is an active and relatively new field, with many studies demonstrating promising results. However, there is often no practical way to apply these findings in real-world settings. The ATS described in this paper utilizes a generic affective application model to infer and appropriately respond to the learner's affective state. This approach brings several advantages, notably the potential direct support for re-use and retrospective addition of affect sensing functionality into existing e-learning software. Skin conductivity and heart rate variability measurements were used to infer affective activation and valence. The evaluation involved an experiment in which the effectiveness of the fully functional ATS was compared with that of a non-affective version, and was conducted with 40 adult participants. The evaluation of the effectiveness of this tutoring system showed that measurable improvements in perceived learning may be obtained with a modest level of software development.