Recommendation systems are used to improve the convenience and efficiency for users tobook hotels. The most widely used method in recommendation systems is collaborativefiltering. A critical step of the collaborative filtering method is to analyze one user'spreference and recommend products or services to the user based on other similar users'preferences. However, collaborative filtering is vulnerable for recommendation when it isdifficult to obtain user preferences, in the situation where e.g. a user provides none or veryfew comments on products or services. The problem occurring in this situation is called thecold start problem. This thesis proposes an improved method which combines collaborativefiltering with data classification to recommend suitable hotels to new users. The accuracy ofthe recommendation is determined by the rankings so that evaluations are conducted on theTop-3 and the Top-10 recommendation lists using the 10-fold cross-validation method andROC curves. The results show that the Top-3 hotel recommendation list proposed by thecombined method has the superiority of the recommendation performance than the Top-10 listunder the cold start condition in most of the times.