The method of recursive discrete wavelet noise filtering for improving metrological characteristics of measuring instruments was investigated for the first time. Methods with a common threshold for all decomposition levels, methods without threshold with a simple zeroing of detail coefficients until the minimum mean square (RMS) error is reached, and methods with universal threshold for detail coefficients at each decomposition level were studied. Twenty different types of measurement signals from the popular PyWavelets library were analyzed. The functions of filtering methods with a common threshold were determined, for which the use of recursion reduces the filtering error from 10 to 50%. For methods without threshold and with universal threshold, the recursion does not reduces the error by multiple filtering of measurement signals. To apply the recursion to the method with a common threshold for all decomposition levels, a mathematical model based on the fundamental equations of wavelet filtering was constructed. The character of distribution of the filtering RMS error depending on the number of reversible cycles is investigated. It was summarized that for the measurement signal models under consideration, the maximum error reduction occurs between the zero cycle, in which the initial measurement signal is filtered, and the first level of recursion. Further reduction of the filtering error with increasing number of recursion cycles occurs according to the law close to hyperbolic.