Objective To analyze risk factors of malignancy in patients with small pulmonary nodules （diameter ≤2 cm） using univariate analysis and multivariate logistic regression，and establish a mathematical prediction model to estimatethe probability of malignancy. Methods Clinical data of 147 patients with small pulmonary nodules who underwentsurgical resection with definite postoperative pathological diagnosis from January 2005 to September 2012 in the 161st Central Hospital of PLA were retrospectively analyzed. There were 84 male and 63 female patients with their age of 31-78（56.2±10.1） years. Univariate analysis using Chi-square test or t test was performed to analyze risk factors including patientage，gender，symptoms，history and quantity of smoking，history of heavy drinking，history of tumor，tumor site，diameter，lobulation，spiculation，pleural indentation，ground-glass opacity，cavity，enlarged hilar and mediastinal lymph nodes.Independent predictors of malignancy were screened with multivariate logistic regression analysis. A mathematical predictionmodel was built to estimate the probability of malignancy and then examined. Results Univariate analysis showed that there was statistical difference in patient age（t=7.146，P＜0.001），heavy smoking history（χ2=6.169，P=0.013），nodule diameter（t=3.375，P=0.001），spiculation（χ2=5.609，P=0.018），lobulation（χ2=5.675，P=0.017），and pleural indentation（χ2=12.994，P＜0.001）between benign and malignant small pulmonary nodule groups. Multivariate logistic regression analysis showed that patient age （OR=1.110，P=0.000），nodule diameter （OR=2.050，P=0.029），lobulation （OR=1.672，P=0.045），spiculation（OR=2.054，P=0.032） and pleural indentation（OR=4.090，P=0.024）were independent predictors of malignancy in patients with small pulmonary nodules （P＜0.05） . The mathematical prediction model to estimate the probability of malignancy was：Logit （P） =ez/ （1 ＋ ez），Z=-6.657 ＋ （0.104×age） ＋ （0.718×diameter） ＋ （0.720×spiculation） ＋（0.514×lobulation） ＋ （1.409×pleural indentation），and e was natural logarithm. Both Hosmer-Lemeshow test （χ2=1.802，P=0.986） and maximum likelihood ratio test （Cox-Snell R2=0.310，Nagelkerke R2=0.443） showed satisfactory goodness of fit. The diagnostic accuracy was 85.71%，sensitivity was 87.50%，specificity was 81.40%，positive predictive value was 91.92%，and negative predictive value was 72.92% when the cut-off value was 0.58. Conclusions Patient age，nodule diameter，spiculation，lobulation and pleural indentation are independent predictors of malignancy in patients with small pulmonary nodules. The mathematical prediction model can accurately estimate the probability of malignancy for patients with small pulmonary nodules.