Passive focusing techniques, which are based on the analysis of image sequences to decide the optimal imaging plane position and moving direction for focusing, are widely used for stroboscopic autofocus system nowadays. When a CCD camera is used to get images, noise of camera may result in errors in the images, and cause the autofocus process inefficient. Thus it is necessary to study microscopic images affected by camera noises to choose the suitable autofocus method. Camera noises usually include photon noise, dark current noise, photo response non-uniformity noise and read-out noise. In our CCD camera-microscopic experiment system, a micro accelerometer is used as the measured object. To analyze the effect of camera noises, we add some camera noises to the images, and run the autofocus process. One of the key techniques for microscopic autofous is its criteria function. Criteria function curves of images are analyzed and compared, including image gradient energy function, variance function, Tennengrad function and Laplace function. Spatial Poisson distribution noises and spatial Gaussian distribution noises are added to images to simulate the camera noises with different expectations and variances. Experiment results show that the performances of these criteria functions are different with different noises added.