This paper presents a direct model reference approach to the off-line design of linear controllers, suited to deal with plants described by a single set of open-loop I/O measurements only. The method is direct inasmuch as the controller parameters are directly estimated with no preliminary identification of any model to describe the plant. The design can be carried out off-line and, in the present formulation, lends to a nonadaptive controller. The basic idea is that of interpreting the open-loop I/O measurements of the plant as closed-loop data produced by a "virtual" reference signal that can be computed by backpropagating the measured output of the plant through the reference model; thus, the controller design reduces to a standard identification problem, in which the "output" signal to be matched is the measured input of the plant. Both a deterministic (noise free) and a stochastic setting are considered.