We study bilinear matrix regression with matrix-valued inputs and outputs, propose optimization-free estimators, derive non-asymptotic error bounds, and validate performance on synthetic and image data.