We consider retrial queues with servers that are subject to active breakdowns. First, we concentrate on Markovian queueing models. For the multiserver case we obtain sufficient conditions for ergodicity. For the singleserver case we introduce new performance measures: the “orbit” idle period and the “orbit” busy period. Further, we investigate the asymptotic behaviour under high intensity of retrials. We also give an algorithm to compute the steady‐state probabilities for theM/G/lretrial queue with breakdowns. The algorithm is based on a regenerative