The high-accuracy nuclear datum of prompt fission neutron spectrum (PFNS) is not only an important parameter for evaluating nuclear data, but also relevant to a more fundamental understanding of the fission process. However, the PFNS experimental data of main actinides (uranium and plutonium) are very scarce and the existing experimental data are in significant discrepancy. The error of some experimental data is markedly large:the relative uncertainty value reaches above 30%. In order to clarify these discrepancies and reduce the uncertainty, more reliable and accurate fission spectrum measurements are necessary. Most of the traditional measurements of the PFNS rely on time-of-flight technique. In this technique the fission sample needs to be prepared in a fission chamber and the fission neutron is identified by fission fragment. Since the particle range of fission fragment is very short, the thickness of sample is limited. So the number of samples cannot be large and the statistic count of fission neutrons will not be great either. Now a new technique to measure PFNS has appeared based on fission gamma multiplicity, which could be used to discriminate fission neutron from other neutrons. The new technique is based on a physical fact:when a fission event happens, seven-to-eight associated gamma photons are emitted while the inelastic scattering effect emits only one or two associated gamma photons. Accordingly, the fission neutron could be discriminated from other neutrons by fission gamma multiplicity. The principles of the new method and the realizing approach are presented in detail. The experimentally measuring system is established. A spontaneous fission neutron source 252Cf is used to measure the PFNS in order to validate the measuring system. The measured spectra are compared with those of ENDF/B-VⅡ library. The PFNS of 238U induced by D-T neutron is measured by the measuring system. The results show that the new method based on fission gamma identification is be available in the measuring of PFNS. The identification efficiency, the total uncertainty analysis of the new method and the suggestion for improvements in the future are also included in the paper.