CO and CH4 as the fault gases in transformer, the detection of their concentration has important significant in transformer maintenance. In order to detect CH4 and CO gas concentration in transformer accurately, a photoaoustic spectroscopy multi-component gas detection system based on broadband thermal radiation light source has been developed. Compared to resonant photoacoustic cell based photoacoustic detection system, the non resonant photoacoustic cell used in the system has the advantage of small volume, easy to processed, high sensitivity and the intensity of signal is the same everywhere in non-resonant photoacoustic cell, thus reduce the requirements for the installation of the detector. The performance of the system was evaluated by detection of CH4 and CO. First, the relationship of the intensity of photoacoustic signal in non resonant photoacoustic cell between the radius and the modulation frequency was simulated by theoretical, it is illustrated that the signal enhanced with the reduction of the radius and the modulation frequency. Then the optimum modulation frequency of the system was determined as 22 Hz by the response of the system under different modulation frequencies. The function relations of photoacoustic signal and the temperature were studied, then the photocoustic signal was corrected by temperature compensate based on the function of signal and gas temperature under optimum modulation frequency to eliminate the impaction of temperature to photoacoustic signal, the stability of the system was improved after temperature compensate. The drift of the signal to temperature is 0.023 23V/℃ and 8.383 48×10－5 V/℃ respectively before and after temperature compensate. At last, the system was calibrated by different concentrations of CH4 and CO gas, the experiment show that the photoacoustic signal increased with the increasing of gas concentration and the linearity can reach to 0.995 and 0.998 4 respectively for the detection of different concentrations of CH4 and CO gas. The multi-component gas detection system based on broadband photoacoustic spectroscopy has the sensitivity of 1μL/L for CO and CH4 detection under atmospheric pressure with the 1s integration time. The developed system has low cost, good linearity and the sensitivity conform to the requirement of national standard in the process of transformer maintenance.