The choice of parameters for procedural textures to achieve a desired appearance poses a challenging problem even for experienced artists. We propose a method to automatically determine such parameters to reproduce the appearance of input images. Addressing two-tone textures, we separate the estimation of color and structure information and interpret the problem as image retrieval task from the space of procedural outputs. Applying a perceptually motivated image metric based on a texture descriptor enables us to precompute a comprehensive collection of possible parameter sets and yet achieve interactive retrieval performance. Our method supports a large variety of procedural texture models with a unified approach.