Searching for the optimal shutter sequence is the key problem in coded exposure photography. Previous shutter sequence searching methods focus on the point spread function estimation and invertibility, and ignore the influence of the scene light level or avoid noise calibration of real cameras. For practical purposes, we address the problem of finding an optimal shutter sequence for coded exposure photography in the presence of photon noise. We analyze the effect of photon noise on the optimal shutter sequence in terms of deconvolution noise and derive analytic formulas. We show that Raskar's code is a special case of our analysis. Based on noise calibration of the coded exposure camera, an effective fitness function is proposed, and using our carefully designed genetic algorithm, we obtain the optimal shutter sequence with little running time. Experimental results with synthetic and real data demonstrate the advantage of our approach compared to the state of the art approach.