This paper presents results of reducing color patches printed during a calibration using a variable selection approach for color electrophotography (EP). During a calibration, primary color patches at different halftone levels are printed on a belt and measured using on-board sensors. Regression models are used to predict primary color tone values on output media from these onboard sensor measurements. Then necessary adjustments using tone correction or other control variables can be applied to compensate tone deviation. Laying down and measuring color patches consumes time and toner. It is desired to determine a minimal set of color patches for the regression models while ensuring their prediction accuracy. This work proposes a procedure that reduces color patches in regression model development. The procedure applies variable selection algorithms to identify the appropriate number of color patches and their associated halftone levels. Results demonstrate that the proposed method can use 45% fewer color patches for the regression models while maintaining prediction accuracy effectively the same as that of models using all the available color patches.