Patients with Parkinson’s disease (PD) show eye movement abnormalities, including slow tracking speed and reduced tracking accuracy. Smooth pursuit eye movement tasks of appropriate difficulty and specific eye movement features can assist in the early diagnosis of PD to formulate prevention and treatment strategies. Our purpose is to evaluate the effects of speed, duration and task difficulty on the eye movement performance of PD and healthy individuals by setting up smooth pursuit tasks with different speeds and complexities. Mann-Whitney U test and Independent sample T test are used to analyze the difference of the eye movement features between healthy individuals and PD patients in three tasks: fast straight line, slow straight line and fast triangular line. By selecting specific eye movement features, we classify PD and healthy individuals based on Support Vector Machine, Random Forest, K-Nearest Neighbor and Logistic Regression classifiers. The results show that the fast triangular line task has the highest classification accuracy of 89.19%, and the slow straight line task has the lowest classification accuracy of 62.42%. The classification results correspond to the results of the statistical analysis. The experimental results demonstrate that PD can be effectively diagnosed by setting up a smooth pursuit task with appropriate task difficulty and selecting appropriate eye movement features.