The applicability is explored of the Bayesian random-effect analysis of variance (ANOVA) model developed by G. C. Tiao and W. Y. Tan (1966) and a method suggested by H. K. Suen and P. S. Lee (1987) for the generalizability analysis of autocorrelated data. According to Tiao and Tan, if time series data could be described as a first-order autoregressive series with parameter "p" (rho), unbiased estimates of random error variance could be derived via a Bayesian process. Suen and Lee's two-step alternative procedure combines both Box-Jenkins time series analysis and a random-effect ANOVA process. The autocorrelated component of the data can be removed through the Box-Jenkins procedure, and then the residual or white-noise data can by analyzed via the ANOVA process to produce unbiased variance estimates. Theoretical advantages and limitations of the two approaches are outlined, focusing on autoregressive integrated moving averages. Three analyses of the methods are presented. Results from application of the methods to numerous data sets show that autocorrelation has a negligible or no effect on the systematic variance across observers. The Suen-Lee method is superior to the Tiao-Tan method in applications to generalizability assessment of observation data. Based on 28 behavioral observation time series, the Suen-Lee method seems to be applicable only when the relative systematic observer variance is small. A 29-item list of references and one data table are provided. (TJH)