This talk will describe a multidimensional (MD) approach to recommender systems that provides recommendations based on additional contextual information, such as the date, time and place when a recommendation is made, besides the traditional dimensions of users and items used in most of the current recommender systems. In addition to multiple dimensions, this approach also supports comprehensive profiling, hierarchical aggregation (OLAP) and querying capabilities. One of the central problems in recommender systems is to devise functions that accurately estimate unknown ratings based on the known ratings previously assigned by users to items. In this talk, different types of multidimensional rating estimation functions will be described, including reduction-, model- and heuristic-based functions. Then the talk will focus on the reduction-based rating estimation function and will compare its performance with the standard collaborative filtering method for estimating unknown ratings. It will be shown that in certain circumstances the reduction-based method outperforms the traditional collaborative filtering method and underperforms it in other circumstances. To leverage the strengths of both methods, a hybrid approach will be presented that identifies the situations where the multidimensional approach outperforms the standard collaborative filtering approach and uses it in those situations and the collaborative filtering approach elsewhere. Finally, a pilot empirical study of recommending movies to people using the multidimensional approach that incorporates such additional contextual dimensions as time, place and companions, will be described. The predictive performance of the reduction-based, traditional collaborative filtering and the hybrid approaches will also be compared as a part of this study.