The oscillation frequency of a combustion flame that is directly measured from the spectrum of a flame signal often depends on the level of white noise in the signal and the sampling rate. As long as the signal contains white noise, the "true" value of the oscillation frequency will never be known. However, the effect of the white noise and sampling rate on oscillation frequency measurement can be minimized by incorporating a new computing algorithm into the measurement software. The algorithm is developed by making two improvements to the original direct calculation method. One of the improvements is the wavelet-based prefiltering of the raw flame signal. The other is an adaptive truncation of the spectrum of the filtered signal. The improved algorithm is evaluated by using a flame signal that is acquired from an industrial-scale combustion test facility. Results that are obtained are reported and discussed.