There are several methods to halftone continuous image. The widespread methods are dithering and error-diffusion. Non-linear color transform by using neural network depends on halftoning methods. Error diffusion has jump of tone in the middle tone area. This affects teaming of neural network and results in increasing color difference. In this paper, we apply FM screen based on visual characteristic to color image. Tone density curve of FM screen is better than that of error diffusion. As a result, using FM screen as halftoning method affects colorimetric color reproduction by neural network and results in decreasing color difference.