This paper addresses the problem of constructing suitable metrics for the evaluation of computer system heterogeneity. There are several metrics available in the open literature that evaluates the heterogeneity degree of a given system. However, this paper shows that many of these available metrics are not entirely suitable when used to support both load and process scheduling mechanisms. It is shown in this paper that some of the available metrics have a dual behaviour for the same system, depending on the adopted reference. A novel metric, constructed from a new approach using the standard deviation concept, is proposed in this paper. This metric is shown to be adequate for all the case studies adopted and it has potential to support most of the load and process scheduling mechanisms used in parallel/distributed computing. The paper also presents results from many case studies considering both the existing metrics and the new one.