A mathematical method has been developed which reduces nuclear-reactor gamma pulse-height spectral data to the form of differential energy spectra through the use of a high-speed computer machine. In essence, the method consists of a least-squares fit of weighted multichannel analyzer data and the utilization of curve-smoothing parametrization. The least-squares approach tends to reduce the magnitude of data that must be handled, i.e., reduces the order of matrix involved. Weighting is used to obtain fractional deviations for minimization by lease squares and thus obtain a satisfactory fit throughout the entire channel range. The parametrization smoothes the reduced data by making use of the fact that reactor gamma spectra can be represented by the product of an exponential and a polynomial. Difficulties that arise when pure matrix inversion is applied have been obviated, and the advantage of high-speed data reduction is gained through the use of an IBM 704-7090 computer program. Error analyses have been undertaken, and data have been reduced for comparative purposes. Results are included in the presentation of the investigation.