Stir-heating is a common method to process Gardenia jasminoides Ellis. The processing of the G. jasminoides Ellis which do not meet the regulated requirements can result in the unqualified products, finally decreasing the efficacy of the drug. To guarantee the quality of the stir-baked G. jasminoides Ellis, a rapid and efficient method to discriminate the qualified stir-baked G. jasminoides Ellis from the unqualified was developed based on the middle level data fusion. 61 batches of qualified and unqualified stir-baked G. jasminoides Ellis were collected and analyzed using colorimeter and near-infrared spectroscopy, separately. L*, A*, and B* were apparent features that quantitatively described the characteristics of samples. The latent variables of partial least squares discriminant analysis model were extracted as the features of near-infrared spectroscopy. The extracted features from colorimeter and near-infrared spectroscopy were fused and a discriminant model was built using partial least squares discriminant analysis method. Accuracy, sensitivity, and specificity were used to evaluate the ability of classification. As a result, the accuracy, sensitivity, and specificity of the optimal partial least squares discriminant analysis model based on fused data were all equal to 1. It demonstrated that the developed method can be capable of discriminating the qualified stir-baked G. jasminoides Ellis from the unqualified. Compared with classification model results using color parameters or near infrared data alone, this data fusion strategy greatly improved the accuracy of classification.