Abstract Study question Is there any intra-surgical parameter able to predict the outcome of salvage microdissection testicular sperm extraction (mTESE) in patients with previous failed TESE? Summary answer Among all the variables under consideration, only the seminiferous tubules (ST) caliber pattern found at high magnification was able to significantly predict the mTESE outcome... What is known already Several studies have demonstrated that no clinical or hormonal parameters are able to predict the outcome of a salvage mTESE performed in patients with previous sperm retrieval failure (SRF). It has been previously demonstrated that a prediction model with the combination of two intra-surgical parameters such as the STs caliber, defined as dilated tubule (DT), slightly dilated tubules (SDT) and not dilated tubules (NDT), and testis histology had an excellent discrimination ability (AUC 0.93) to distinguish between cases with and without the outcome, but such prediction model has not been tested in patients undergoing salvage mTESE. Study design, size, duration A prediction model was built on a dataset of 63 patients, 29–50 years old, undergoing unilateral (15 ) or bilateral (48) salvage mTESE after failed TESE from 2015 through 2019, with a resulting N = 111 testes under consideration. Two models were compared, one with STs and histology as covariates, the other with STs alone: the second model was chosen due to better discrimination... Participants/materials, setting, methods we assessed internal validity with a bootstrapping procedure for a realistic estimate of the performance of the prediction model in similar future patients with NOA undergoing salvage mTESE. We repeated the entire modeling process in 259 samples drawn with replacement from the original sample, and determined the performances (AUC, sensitivity, specificity) of the selected prediction model. Calibration (correspondence between the predicted and observed probabilities) was visually assessed by inspecting the calibration belt... Main results and the role of chance Sperm retrieval was successful in 24 out of 63 patients (38%): age, testis volume and hormonal parameters did not vary among patients with successful sperm retrieval (SSR) or SRF. The prevalent histological pattern was Sertoli cell only syndrome (69.6%), while hypospermatogenesis, maturation arrest and hyalinosis were found in 4.5%, 23% and 1.8% of cases. The STs pattern was heterogeneous, with DTs being found only in 23.4% of testes. Sperm were found in 69% of DTs, 29% of SDTs, and 5% of NDTs. The prediction model correctly classified 82.88% of patients and explained the 26.5% variability of the outcome. The STs pattern significantly predicted the mTESE outcome with a sensitivity of 62% and a specificity of 90.2%, PPV 69.2%, NPV 87%. Both SDT (OR 0.105, 95% CI 0.034–0.317, p < 0.0001) and NDT (OR 0.024, 95% CI 0.004–0.128, p < 0.0001) were negatively associated with the chance of retrieving sperm, the resulting prediction equation being Log (SSR)= 0.81 – 2.2 SDT – 3.7 NDT. The model had a clearly useful discrimination (AUC 0.813). The optimism corrected AUC was 0.7977, and the model was well calibrated (p = 1.00) with both the 80% and 95% calibration belts encompassing the bisector over the whole range of the predicted probabilities Limitations, reasons for caution The STs caliber pattern was subjectively evaluated at high magnification (24–36x) by comparing the individual ST appearance with the surrounding ones, however such evaluation was performed by the same experienced urologist with more than 1000 mTESE procedures performed to date. Wider implications of the findings: No clinical data but the STs appearance at high magnification could discriminate between patients with and without chances of SSR. These results reinforce the evidence supporting the superiority of mTESE compared to conventional TESE in retrieving sperm, particularly in lower prognosis patients with NOA such as those with previous SR. Trial registration number Not applicable