Can a counselling tool be developed for couples with different types of azoospermia to predict the probability of clinical pregnancy in ICSI after surgical sperm retrieval? A prediction model for clinical pregnancy in ICSI after surgical sperm retrieval in different types of azoospermia was created and clinical type of azoospermia, testicular size, male FSH, male LH, male testosterone, female age, female antral follicle count (AFC) and female anti-Müllerian hormone (AMH) were used as predictors. Prediction models are used frequently to predict treatment success in reproductive medicine; however, there are few prediction models only for azoospermia couples who intend to conceive through surgical sperm retrieval and ICSI. Furthermore, no specific clinical types of azoospermia have been reported as predictors. A cohort study of 453 couples undergoing ICSI was conducted between 2016 and 2019 in an academic teaching hospital. Couples undergoing ICSI with surgically retrieved sperm were included, with 302 couples included in the development set and 151 couples included in the validation set. We constructed a prediction model using multivariable logistic regression analysis. The internal validation was based on discrimination and calibration. We found that for male patients involved in our model, different clinical types of azoospermia are associated with different clinical pregnancy outcomes after ICSI. Considering the clinical type of azoospermia, larger testicular volume and higher levels of FSH, LH and testosterone in the body are associated with higher clinical pregnancy success rates. For women involved in our model, younger age and higher AFC and AMH levels are associated with higher clinical pregnancy success rates. In the development set, the AUC was 0.891 (95% CI 0.849-0.934), indicating that the model had good discrimination. The slope of the calibration plot was 1.020 (95% CI 0.899-1.142) and the intercept of the calibration plot was -0.015 (95% CI -0.112 to 0.082), indicating that the model was well-calibrated. From the validation set, the model had good discriminative capacity (AUC 0.866, 95% CI 0.808-0.924) and calibrated well, with a slope of 1.015 (95% CI 0.790-1.239) and an intercept of -0.014 (95% CI -0.180 to 0.152) in the calibration plot. We found that BMI was not an effective indicator for predicting clinical pregnancy, which was inconsistent with some other studies. We lacked data about the predictors that reflected sperm characteristics, therefore, we included the clinical type of azoospermia instead as a predictor because it is related to sperm quality. We found that almost all patients did not have regular alcohol consumption, so we did not use alcohol consumption as a possible predictor, because of insufficient data on drinking habits. We acknowledge that our development set might not be a perfect representation of the population, although this is a common limitation that researchers often encounter when developing prediction models. The number of non-obstructive azoospermia patients that we could include in the analysis was limited due to the success rate of surgical sperm retrieval, although this did not affect the establishment and validation of our model. Finally, this prediction model was developed in a single centre. Although our model was validated in an independent dataset from our centre, validation for different clinical populations belonging to other centres is required before it can be exported. This model enables the differentiation between couples with a low or high chance of reaching a clinical pregnancy through ICSI after surgical sperm retrieval. As such it can provide couples dealing with azoospermia a new approach to help them choose between surgical sperm retrieval with ICSI and the use of donor sperm. This work was supported by a grant from the National Natural Science Foundations of China (81501246 and 81501020 and 81671443). The authors declare no competing interest. N/A.