In the field of cell biology, particle counting in intracellular images is important for investigating the cause of diseases. However, particles are manually counted by human observers now. Such manually counting takes a lot of time, and counting result becomes subjective. If an automatic counting method by computer is realized, we can treat a large number of images and it can obtain many objective data. The development of an automatic counting method much contributes to understand the case of disease. However, particle counting in intracellular images by a computer is new research field, and conventional methods are little. Thus, we propose counting method based on regression analysis. We use partial least squares regression and auto-correlation between 2 different types of features using mask patterns for higher-order local autocorrelation feature. The proposed method gives higher accuracy than counting by principal component regression, support vector regression and ImageJ. In experiment, the proposed method can count with small error in comparison with human counting. The effectiveness of our method is shown by experiments.