The paper presents optimization of kernel methods in the task of handwritten digits identification. Because such digits can be written in various ways (depending on the person’s individual characteristics), the task is difficult (subsequent categories often overlap). Therefore, the application of kernel methods, such as SVM (Support Vector Machines), is justified. Experiments consist in implementing multiple kernels and optimizing their parameters. The Monte Carlo method was used to optimize kernel parameters. It turned out to be a simple and fast method, compared to other optimization algorithms. Presented results cover the dependency between the classification accuracy and the type and parameters of selected kernel.