Past recommendations for analysis of the multitrait-multimethod matrix have emphasized validation rather than traits and measures development. This paper emphasizes the latter, and a review of existing approaches suggests that each has shortcomings for developmental analysis of the matrix. Structured types of analysis, with pre-hypothesized models, are often too cumbersome, do not allow efficient examination of trait-method interaction or, most important, hold the potential of false negatives with too early rejection of traits and measures. Unstructured approaches typically use factor analysis, are restricted by its linear and parametric data requirements, and hold the potential of false positives with too easy acceptance or formation of traits. A combination of cluster analysis and nonmetric scaling offers the significant advantages of nonlinear and nonparametric data input, a sequence of levels of analysis, clear visual depiction of the matrix, reproducibility, and cross checking between two analysis approaches. These advantages are illustrated in a comparison of analysis approaches as applied to a “successful” matrix of clinical assessment data and a “weak, developmental” matrix from a political attitudes study. The clustering-nonmetric scaling approaches confirm the results of previous analyses but offer additional information for trait development.