Prior studies have observed that exposure variability increased as a function of sampling duration and attributed this phenomenon to autocorrelation. This study confirmed such behavior in occupational exposure data after controlling for factors likely to contribute to variability and assessed the impact of non-stationarity, as well as autocorrelation, on the results. Consecutive shift-long exposure measurements for 54 workers from five different data sets in 149 time series were analyzed to evaluate the variance as the interval between measurements increased. When the data were combined a clear increasing trend in the variance was observed with lag. However, a breakdown by data set revealed that the trend was present in only one of the five data sets. The effect was further isolated to 42% of the workers who contributed data and to less than 1/3 of the total number of time series analyzed. Autocorrelation and non-stationary behavior explained the increase in 60% of the time series where the trend was evident. Analysis of the entire database revealed that a small percentage of time series produced significant first-order autocorrelation coefficients or were non-stationary over the interval in which sampling was conducted. If these results are typical of other workplaces, sampling strategies may not need to address problems associated with autocorrelation or nonstationarity.