Robert C. Martin presented a software metric for a set of classes i.e. a package. The objective of the package level metric is to identify poorly designed packages. The Martin's metric actually consists of eight metrics which measure a few different characteristics of packages. The metric is widely known, but there is lack of theoretical and empirical evaluation of the Martin’s metric. This paper evaluates the theoretical background of the metric against an evaluation framework and presents an experimental evaluation of five open-source software applications. The theoretical validation reveals a weakness in Martin's definition for cohesion. We propose a modification which is valid according to the evaluation framework.