Objective The detection of ink film thickness of metal sheet printing is one of the key technologies in the printing quality control field. According to the development trend and the characteristics of metal sheet printing, an online detection system was proposed for the ink film thickness of metal sheet printing. Methods First, the ink film thickness d and real-time image's RGB values of the signal bars were obtained through experiments. Then, the relationships of the basic attributes of colors (lightness (L*), saturation (C*ab) and hue (hab)) calculated from the RGB values and the thickness of the ink film were analyzed and the significant corresponding relationships among them were found. Last, an ink film thickness prediction model was established using Extreme Learning Machine (ELM). Results The simulation results showed that the accuracy of the ELM method was much higher than BP and RBF methods with the average relative error of 2.27%, and the maximum relative error of 7.33%. Conclusion It was proved that this detection system could well realize the function of ink film thickness detection and could be applied to the actual testing process.