Abstract: Derivation of neutron energy spectrum from multi-sphere neutron spectrometer’s experimental data is a kind of few channel problems, and therefore has more than one solution. Most unfolding methods try to search among the solution space to find the solution that best fit the measurement data and the response function. As a kind of optimization strategy, genetic algorithm could find the global optimal among the search space. Anew fitness function which contains a distance part and a penalty part was constructed in this research. The distance part is the square distance between the individual and the measurement data. The penalty part which is a function associated with the continuityof individual was used to avoid intensively change of unfolded data. Five classical neutron spectra were chosen as benchmark spectra. Theresults of the benchmark spectra multiplied by the response function were acted as input measurement data of the unfolding program.Unfolded results show that they are well agreeable with the true spectra, proven the feasibility of unfolding neutron spectrum using genetic algorithm.