During the last years, a new technique called ICA (Independent Component Analysis) has attracted considerable attention. The aim of ICA is, given a number of observed linear combinations of some unobserved time series, to estimate the unobserved time series. The estimation is performed only on the basis that the time series are mutually statistically independent, thereof the name ICA. It would be interesting to find out how ICA could be used in a sonar system to separate different sounds from each other, or to separate desirable sounds from noise. In this thesis a literature study, which describes ICA and results from some scientific articles that discuss the separation of sound sources, is presented as an introduction. After that, a method for a special case where two sensors, interspaced a few meters, simultaneously receives the sound from a transmitter whose signal has known statistical properties, and the sound from an interfering source, is proposed. With the aid of a method that is similar to ICA, the desired signal is separated from the interfering sound. An evaluation of the method shows that the method works well under idealized conditions, but is sensitive to deviations from the mathematical model on which the method relies.