AbstractA method for estimating the drift function and assessing the significance of possible trends in spatial observations was developed. This method takes as many data points as possible within each lag distance, is less sensitive to “outliers,” and nearly all observations obtained by it have their contribution in calculating the drift function. Our method, as well as the “cancellation method,” were examined on temperature data collected at a depth of 0.05 m at 0.68‐m intervals along an 80‐m long transect. Tests based on the cancellation method falsely indicated the presence of a significant trend in the temperature data. No trend was indicated by our method, which has more statistical validity than the cancellation method. The decision to detrend or not has a marked impact on the interpretation of the data as attested to by the semivariograms and autocorrelograms of the original and detrended data. Detrending of the data resulted in a loss of information about the spatial dependence of temperature observations.