Daily diaries of respiratory symptoms are a powerful technique for detecting acute effects of air pollution exposure. While conceptually simple, these diary studies can be difficult to analyze. The daily symptom rates are highly correlated, even after adjustment for covariates, and this lack of independence must be considered in the analysis. Possible approaches include the use of incidence instead of prevalence rates and autoregressive models. Heterogeneity among subjects also induces dependencies in the data. These can be addressed by stratification and by two-stage models such as those developed by Korn and Whittemore. These approaches have been applied to two data sets: a cohort of school children participating in the Harvard Six Cities Study and a cohort of student nurses in Los Angeles. Both data sets provide evidence of autocorrelation and heterogeneity. Controlling for autocorrelation corrects the precision estimates, and because diary data are usually positively autocorrelated, this leads to larger variance estimates. Controlling for heterogeneity among subjects appears to increase the effect sizes for air pollution exposure. Preliminary results indicate associations between sulfur dioxide and cough incidence in children and between nitrogen dioxide and phlegm incidence in student nurses.