The problem of top-k skyline computation has attracted considerable research attention in the past few years. Given a dataset, a top-k skyline returns k "most interesting" skyline tuples based on some kind of preference specified by the user. We extend the concept of top-k skyline to a so-called top-k combinatorial skyline query (k-CSQ). In contrast to the existing top-k skyline query (which is mainly to find the interesting skyline tuples), a k-CSQ is to find the interesting skyline tuples from various kinds of combinations of the given tuples. The k-CSQ is an important tool for areas such as decision making, market analysis, business planning, and quantitative economics research. In this paper, we will formally define this new problem, propose an intelligent method to resolve this problem, and also conduct a set of experiments to show the effectiveness and efficiency of the proposed algorithm.