The invention provides a method for collaborative filtering recommendation based on item level types. The method for collaborative filtering recommendation based on the item level types comprises the four steps of preference description, item type grading, similarity calculation and predictive grading. In the step of preference description, modeling of preference of a user is completed and a user-item grading matrix is generated. In the step of item type grading, grades, offered by a user, of bought items are defined, the similarity between the item types is defined according to the association rules, and grades, offered by the user, of item types which are not bought are deduced. In the step of similarity calculation, the similarity between every two items of a system is obtained according to the similarity calculation formula. In the step of predictive grading, grades, offered by the user, of the items which are not graded are predicated. According to the method for collaborative filtering recommendation based on the item level types, the number of item level type factors is increased, the new similarity formula is established, the influence of sparseness of the user-item grading matrix on the accuracy of the item similarity is reduced, the probability of the phenomenon that the similarity between every two items is zero is reduced, and the accuracy of recommendation is remarkably improved. The method for collaborative filtering recommendation based on the item level types can be applied to the fields such as the field of data miming and the field of recommendation systems.