Developmental psychologists have not traditionally documented the psychometric properties of their observational instruments, partly because such concepts as reliability and validity are associated mainly with paper-and-pencil tests. A multitrait-multimethod correlation matrix (MTMM) was presented as a useful method for the psychometric evaluation of these instruments; two hypothesized constructs important in early social development (fear of strangers and exploration in a novel environment) were measured by a laboratory observation of child behavior, and by a maternal questionnaire. The MTMM applied an evaluation sequence not usually followed by researchers--the sequence consisted of intra- and inter-observer agreement; stability--consistency of children's behavior over a short term; and construct validity (and its prerequisites, convergent and discriminant validity). Observer agreement was a necessary but not sufficient condition for stability, and stability was a necessary but not sufficient condition for validity. (CP)