I applied several established water-quality trend-detection methods to the problem of detecting and quantifying streamflow impacts that could result from the interbasin transfer of water in the New Jersey Pinelands. A major objective of the study was to assess the sensitivity of the different methods in detecting base flow changes. Base flows represent streamflows that are derived entirely from ground water. I subjected base flows at two continuously gaged Pinelands study sites to simulated flow reductions. I then used simple linear regression to relate base flows at these two sites to base flows at five Pinelands index sites generally unaffected by major streamflow impacts. Monotonic-trend (i.e., gradual and continuous trends) and step-trend tests were applied to the residuals from the regression models. Residuals are the predicted study-site base flows subtracted from the actual values. I used the smallest-detectable percentage reduction within a given simulation that was significant at an alpha level of 0.10 as an estimate of the sensitivity of a trend test. I found that the best models were those that displayed the strongest correlation between study-site base flows and index-site base flows while avoiding first-order autocorrelation in the regression residuals. Autocorrelation or serial correlation in regression analysis describes a condition where the residuals are dependent or correlated to each other through time. Autocorrelation arose when the time intervals between samples were short. I found that using a quarterly (3-month) interval rather than a shorter two-month sampling interval reduced the problem of autocorrelation. My analysis also revealed that when data were missing for a particular sampling event, the ability to detect autocorrelation decreased. Trend-test sensitivity appeared to improve when the number of sampling events before and after a base flow impact were equal. Seasonally biased sampling appeared to reduce sensitivity. Based primarily on the results of past trend-detection studies, I recommended using nonparametric tests over their parametric counterparts.