For olfactory sense and gas mixtures identification of humanoid robots, an artificial lung & olfactory sense system (HAL&OS-I) and its identification method through active breathing are proposed and researched. The integrated hardware of the system mainly consists of micro vacuum pump, five gas sensors for alcohol/hydrogen sulfide/ammonia/smoke/methane separately, and the single chip microcomputer along with the circuit boards for signal sampling and processing. Gas identification experiments of five pure gases and four gas mixtures were conducted by using K-mean cluster analysis method, genetic algorithm combined with neural network (GA+BP), cascade neural network (GA+3BP) separately. The experimental results show that the identification rate of five pure gases by the GA+BP algorithm is above 90%, but the identification rate is relatively low when the gas mixtures are included. Gas identification rate of all gases by the GA+3BP algorithm is more than 90% except the smog and hydrogen sulfide mixture gas of which the identification rate is 70%. It is revealed that the GA+nBP algorithm has higher identification rates for multiple pure and gas mixtures.