Spatial rich model (SRM) is one of typical steganalytic features based on rich model and SRMQ1 is its subset. Since they only adopt several quantization step values in constructing features, the performance of steganographic detection is limited. To promote the performance, this paper conducts digital image steganalysis based on spatial rich model features and dimensionality reduction. It first exploits more quantization step values to construct steganalytic features. Then, taking submodel as a unit and adopting Fisher score, Kolmogorov-Smirnov (KS) test and principal component analysis (PCA) to process features of each submodel respectively. Following that, utilizing feature selection based on Fisher score to process steganalytic features and selecting some informative features. Finally, the first 12753 and 34671 features are selected and then compared with SRM and SRMQ1. Experimental results demonstrate that the proposed feature set can achieve better detection accuracy.