〔Abstract〕 203 thyroid micro-nodules pathologically confirmed were retrospectively analysed, including 148 cases of thyroid microcarcinoma and 55 cases of micronodular goiter.The software(AK) was used to perform radiomics analysis on CT plain and enhancement scan images of thyroid micro-nodules. Six models were used for evaluation: Forest model, SVM model, KNN model, Tree model, Bayes model and Logistic model. 70% of the cases were randomly selected as the training group, and the testing group was the remaining 30%. The accuracy, specificity, sensitivity and AUC of different models for the differential diagnosis of two types of micro-thyroid nodules were obtained.The diagnostic efficiency of plain scan and enhanced scan were analyzed and Forest model was the best. The diagnostic accuracy, specificity, sensitivity and AUC of the plain scan were 99.3%, 97.9%,100% and 0.999,respectivelyfor training group while 87.9%, 81%, 91.1% and 0.884 for the testing group. For the enhancement scan, those of the training group were all 100% and the test group were 82.8%, 93.3%, 71.4% and 0.905. For CT radiomics, Forest model is of certain value in the differential diagnosis of benign and malignant thyroid micro-nodule. Plain scan and enhanced CT have high sensitivity and specificity, respectively.