The results of acoustic surveys, especially in low signal-to-noise regimes, are sensitive to the method used to exclude the contribution of background noise from the acoustic data. The use of thresholds to reject weak signals during data collection is inappropriate because the noise contribution to the above-threshold-signal cannot later be removed in post processing. The only solution is to log unthresholded data and to correct for noise in post processing. Integrated data collected at 120 kHz with a Simrad EK500 are presented to demonstrate the highly variable nature of background noise. A dynamic method to determine background noise for individual integration intervals was found to be better than several methods that apply a single background noise level to a data set. The dynamic technique requires that logged data have high resolution in both depth and distance. It can be applied so long as some integration layers in each integration intervals include only noise. Further, it is recommended that the physically meaningful quantity "noise power", being the apparent received power (dB/ 1 W) at the transducer due to noise, is used to describe background noise levels.