Since 1990s, various methods in automatic term recognition (ATR) have been proposed, and there is recently a move towards establishing a gold standard to be used for the evaluation of ATR methods. This presentation discusses some theoretico-technical issues and pitfalls in establishing such gold standards for ATR, partly taking into account our own experience in organising the evaluation-based workshop on ATR and by clarifying the basic points to be considered in the evaluation of the performances of ATR methods. In the process, we also examine some standard quantitative methods for automatic term weighting, such as tfidf, term representativeness, productivity of lexical elements, relating their basic characteristics to the conceptual framework within which different ATR applications can be perceived.