The Mallat algorithm used in wavelet transformation is introduced to determine the deflection point of temperature curve in dew -point auto -detection instruments. The method has several distinguished points: the useful signal from backnoise can be effectively extracted and the highly repeatable value of deflection point can be obtained in only 3 -5 times of measurement. As compared with traditional method. i. e , the fitting curve modelling from the detected data through regressive variance analysis, the method suggested can considerablly reduce the testing time and raise the accuracy and stableness in dew-point auto - measurement.