This paper describes generalizability theory as an extension to the analysis of variance approach to reliability which allows the researcher to attribute variation to multiple sources, and a number of reliability coefficients to be calculated for most observational contexts. The applications of generalizability coefficients to characterize the dependability of data gathered on classroom process variables are pursued. Models are provided for crossed and nested designs for situations involving two and three sources of variation. Several examples are discussed involving data collected with the Teacher Performance Assessment Instruments (TPAI). Other examples discuss applications of generalizability theory to measures of cognitive level of teacher questioning and measures of student engagement. All examples were selected to highlight applications of generalizability theory, which these authors feel is particularly applicable to educational measures which are often affected by a number of identifiable sources of variance. (CS)