We study the phase retrieval (PR) technique using through-focus intensity measurements and explain the dependence of PR on the defocus distance. An optimal measurement plane in the out-of-focus region is identified where the intensity distribution on the optical axis drops to the first minimum after focus. Experimental results confirm the theoretical predictions and are in good agreement with an independent phase measurement. © 2013 Optical Society of America OCIS codes: 220.1080, 120.5050, 100.5070.Phase retrieval (PR) concerns all the types of nonlinear algorithms for recovering phase information when only the intensity of a complex field is known [1][2][3]. PR has widely been applied in the last 20 years to determine the phase aberrations in the exit pupil of an optical system. PR methods that use through-focus (PRTF) intensity images show advantages over standard interferometric techniques [4,5], such as low demand on the temporal coherence requirement and system mechanical stability. Nevertheless, one of the main disadvantages of PRTF is the need for several through-focus measurements, which leads to a heavy computation load. With the recent developments in high-performance computing systems and graphic processing units, PRTF has the potential to be used as a real-time phase measuring technique for application where fast phase measurements are needed, such as adaptive optics systems [6,7] or beam-shaping probing. However, when more than one measurement plane is used to retrieve the phase, uncertainties in the relative distance between two adjacent measurement planes are introduced into the system. In order to minimize the contribution of these uncertainties in the measurement results, it would be attractive to perform the PRTF using only one intensity measurement plane, added to some a priori knowledge of the optical system, such as numerical aperture (NA) and finite support of the pupil [8]. Lee et al.[9] introduced a statistical analysis in order to find the optimal defocus distance based on the Cramer-Rao lower bound. In this Letter, we show that in the limit of small aberrations [10] one can derive an analytical model that identifies the optimum measurement plane to achieve accurate PR. Simulations are carried out, and experimental results verify the predictions. The PR algorithm, considered in this Letter, is based on a nonlinear optimization with respect to the Zernike coefficients [11] that describes the phase-only distribution within the exit pupil of an optical system. A detailed description of the algorithm can be found in [6,12,13]. As an example, we show the performance of this algorithm for four different random phase distributions consisting of 36 Zernike coefficients with the total RMS 0.04λ [ Fig. 1(a)] in an optical system with NA 0