This paper evaluates four metaphor identification systems on the 200,000 word VU Amsterdam Metaphor Corpus, comparing results by genre and by sub-class of metaphor. The paper then compares the rate of agreement between the systems for each genre and sub-class. Each of the identification systems is based, explicitly or implicitly, on a theory of metaphor which hypothesizes that certain properties are essential to metaphor-inlanguage. The goal of this paper is to see what the success or failure of these systems can tell us about the essential properties of metaphorin-language. The success of the identification systems varies significantly across genres and sub-classes of metaphor. At the same time, the different systems achieve similar success rates on each even though they show low agreement among themselves. This is taken to be evidence that there are several sub-types of metaphor-in-language and that the ideal metaphor identification system will first define these sub-types and then model the linguistic properties which can distinguish these sub-types from one another and from nonmetaphors.