By an experimental finding, for a given radiant point source and a given infrared detector sampling, an optical point spread function (PSF) falling on different positions of an infrared detector focal plane array, will result in different measured PSFs. Those measured PSFs known as decoding kernels will further produce different decoded images. Therefore, this work explores the effect of PSF position bias on decoded images in a wavefront coding infrared imaging system. This paper theoretically analyzes the effect of PSF position bias on decoded images. A simulation is conducted for qualitatively evaluating the effect of PSF position bias on decoded images by mean structural similarity (MSSIM) index. Simulated results prove that the probability for MSSIM index greater than 0.75 is 91.38%, and that for less than 0.5 is only 1.22% but this minority seriously degrades the decoded images. We construct an experimental setup for capturing the raw PSF image at a random position and proposes a procedure to reduce the noise of a raw PSF image. The experimental result demonstrates the feasibility for measuring a PSF to achieve a decoded image with good quality.