Sources of errors in mechanical testing attributable to the test instrument and transducers are generally well-understood. ASTM and other standards organizations have well-defined procedures for the calibration of load, strain, and displacement transducers, and the technology for improving the accuracy of these transducers has improved greatly in recent years. However, the calculation of mechanical properties is still highly dependent on the interpretation of stress-strain curves by humans, and the variation in human judgment leads to errors which in many cases are considerably greater than the errors in the transducers used for acquiring stress-strain curves. The problem is further compounded by the fact that ASTM test procedures poorly define mechanical properties, and most of these definitions are qualitative rather than quantitative. The lack of good definitions becomes most apparent when the definitions are translated into numerical algorithms for computerized systems that try to compute mechanical properties without the benefit of human judgment. This paper presents examples of these poor definitions and explains how they can lead to substantial errors in computing mechanical properties even if the stress-strain curve is highly accurate. Definition errors in the calculation of Modulus, Yield Point, Proportional Limit, Break Point, and other commonly used mechanical properties are described. The paper concludes by presenting some computer algorithms to improve the quality and consistency of computing mechanical properties from stress-strain curves.