BACKGROUND: All-cause 30-day readmission after total knee arthroplasty (TKA) is currently used as a measure of hospital performance in the United States and elsewhere. Readmissions from surgical causes may more accurately reflect preventability and costs. However, little is known about whether predictors of each type of readmission differ. METHODS: All primary TKAs recorded in England's National Health Service administrative database from 2006 to 2015 were included. Multilevel logistic regression analysis was used to describe the effects of patient-related factors on 30-day readmission risk using 3 different readmission metrics: all-cause, surgical (defined using International Classification of Disease-10 primary admission diagnoses), and those resulting in return to theater (RTT). RESULTS: In total, 566,323 procedures were recorded. The comorbidity with the highest odds ratio (OR) for all types of readmission was psychoses (RTT OR 2.52, P < .001). Obesity was a strong independent predictor of RTT (OR 1.36, P < .001) and had the highest population attributable fraction of any comorbidity (4.7%). Unicompartmental arthroplasty was associated with a significantly lower risk of all types of readmission when compared with TKA, with the effect being most pronounced for surgical readmission (OR 0.66, P < .001). RTT in the index episode increased the risk of RTT readmission (OR 2.80, P < .001), as did any emergency admission to hospital in the preceding 12 months (for >2 emergency admissions, all-cause OR 2.38, P < .001). Length of stay either more than or less than 2 days was associated with an increased risk of all-cause and surgical readmission but not RTT readmission. CONCLUSION: Patient-related predictors of surgical and RTT readmission following TKA differ from those for all-cause readmission, but only the latter metric is in widespread use.