This document summarizes an analysis of the performance of three named-entity recognition systems: Stanford, LBJ, and IdentifiFinder. The analysis found differences in how each system tokenized text and the total number of entities recognized. It also found ambiguity cases where the same token was assigned different entity types within a single document. To address evaluation issues, the document proposes a standardized unit test with examples of true positive entity types and guidelines for intrinsic evaluation of named-entity recognition.