Knowing the peak to average ratio (crest factor) of the signal a loudspeaker needs to reproduce is important for determining the maximum level a loudspeaker can attain with that signal. An investigation has been made into the crest factor of real signals as a function of frequency. It was found (not very surprisingly) that the crest factor increases with increasing frequency. However, in contrast, it is easily shown that the crest-factor of pink noise (and other standard noise signals) changes minimally with frequency. A new test signal has been created whose crest-factor-as-a-function-of-frequency is modeled after real signals. A procedure is proposed for finding the maximum SPL that a given loudspeaker can reproduce this signal linearly. This signal and procedure has been made publicly available so that comparable measurements can be made by multiple independent individuals. As a signal level is increased and approaches the maximum capacity of the loudspeaker, peaks may be clipped, frequency content may be added that was not in the original signal, and other “non-linear” effects may occur. To measure the onset of non-linearity of a loudspeaker, a test signal must be used which not only has the average spectrum as a function of frequency expected from real signals, but also the crest factor as a function of frequency. Traditional distortion measurements (while valuable) can not be used with sounds that have energy at all frequencies simultaneously. For this reason, the proposed procedure uses coherence. Coherence is a measure of the correlation between the input and output of a system and can be determined from quantities already produced in the course of calculating a transfer function. The transfer function shows the reduction in sensitivity that occurs when the input is increased but the output does not increase proportionally. These two quantities provide a reliably repeatable of the onset of non-linearity and the corresponding maximum SPL.