In gamma-ray spectroscopy, a pulse pile-up occurs when a gamma-ray hits the detector within a time period shorter than the duration of the pulse induced by a preceding gamma-ray. The resulting pulse is the sum of two overlapping pulses. At a high counting rate, the pulse pile-up can lead to significant loss of data and/or deterioration of the measured gamma-ray energy spectrum. If the temporal separation of consecutive gamma-ray events is small enough, the resulting pile-up pulse exhibits only a single peak. Such events are described as peak pile-ups. It is not trivial to detect and correct peak pile-ups. Consequently, such events are typically wrongly reconstructed as a single pulse or rejected by the pile-up correction algorithms developed to date. We study a least-squares fitting algorithm modified to include detection and correction of peak pile-ups. Performance of the proposed algorithm is studied using data measured with LaBr3:Ce scintillator and a 400 MBq 137Cs source as well as with artificially synthesized pile-up data. The pile-up correction success rate grows rapidly with the temporal separation of piled-up pulses and reaches 50% as soon as the pulses are separated by a time at least of the order of the single pulse rise time. A success rate exceeding 90% is reached for the temporal separation equal to about twice the rise time. The reconstruction efficiency measured as the overall rate of the successful reconstruction of the double-pulse pile-up is about 92%.