The mortality of colorectal cancer ranked fifth in China according to cancer statistics in 2015. Cancer screening had been repeatedly proved to play a vital role in decreasing the incidence and mortality of colorectal cancer, but the existing screening methods could not meet the requirements. So it is of urgent need to develop a non-invasive, convenient and accurate screening method.In this study, stool samples were collected from 102 colorectal cancer, 20 colorectal adenoma, 6 hyperplastic polyps patients and 105 normal controls, and stool DNA was extracted for detection of methylation (BMP3, NDRG4, SDC2 and SFRP2) and KRAS mutations. Meanwhile, hemoglobin in stool samples was detected by immunoassays. Then, the logistic regression model used for classification was built with these biomarkers, and a ROC curve was drawn to evaluate the performance of each biomarker and the panel of them. Meanwhile, conventional serum biomarkers were detected for the comparison of positive rate in colorectal cancer between serum biomarkers and stool DNA biomarkers.As a result, a classification model built with methylation of SDC2 and SFRP2, KRAS mutations and hemoglobin showed a sensitivity of 91.4% for colorectal cancer and 60% for adenoma with the specificity of 86.1%. Compared with it, most of the conventional serum biomarkers showed a sensitivity of less than 20% for colorectal cancer which was significantly lower than stool DNA biomarkers.A novel panel comprised of stool DNA biomarkers was of much higher sensitivity and specificity in early screening of colorectal neoplasms than conventional serum biomarkers.