Abstract: A method for learning multivalued mapping comprises the steps of: mathematically expressing a multivalued function directly in Kronecker's tensor product form; developing and replacing the tensor product form so as to obtain a linear equation with respect to unknown functions; defining a sum of a linear combination of local base functions and a linear combination of polynomial bases with respect to the replaced unknown function; and learning or structuring a manifold which is defined by the linearized functions in the input-output space, from example data, through use of a procedure for optimizing the error and the smoothness constraint. Therefore, mapping learning can be performed from a small amount of data.