INTRODUCTION Whether in internally generated thermal reports or analysis， or professional?level publications in technical journals and conferences， use has been made of the concept of “% error in Tj.” Generally speaking， this is totally misguided. This monograph explains why the concept is generally incorrect， the rare circumstances when it may be acceptable， and more importantly， alternatives having superior properties that are more generally applicable. Definitions First off， we must be absolutely clear in what we’re talking about. The “standard” definition of percent error is generally expressed something like：

　　where X is some quantity being judged as to its conformity with a previously established value， expectation， theory， or what have you. The numerator， clearly， is a simple difference between two values; the denominator is the reference value. In the case of junction temperature （and more specifically， in this example we are going to be comparing the result of a “compact thermal model，” or CTM， against a more complicated and detailed finite element model of a package）， we would have the specific definition：