Abstract: The basic principles of wavefront sensing based on phase diversity (PD) were analyzed. An improved PD based on genetic algorithm (GA) was proposed for the multiple variables and nonlinear character of PD's objective function. A Tikhonov regularization function was added to the traditional objective function, and the stability and convergence efficiency were improved. The noise adaptability of wavefront testing by PD technology was simulated, and effect of various wavefront defocuses on the testing precision of PD technology was analyzed. The best wavefront defocus which can make PD reach the highest testing precision, and it was selected to apply in testing experiment. PD technology was applied to wavefront testing of central and marginal field of view in off-axis three-mirror reflecting optical systems. The Zernike coefficients solved and wavefront maps were compared with interferometric results respectively. Experimental results showed that the testing precision of PD technology was less than 0.013 (RMS), and the precision can satisfy the requirement of engineering.