Abstract Drones are being increasingly used to monitor wildlife populations; their large spatial coverage and minimal disturbance make them ideal for use in remote environments where access and time are limited. The methods used to count resulting imagery need consideration as they can be time-consuming and costly. In this study, we used a fixed-wing drone and Beyond Visual Line of Sight flying to create high-resolution imagery and digital surface models (DSMs) of six large king penguin colonies (colony population sizes ranging from 10,671 to 132,577 pairs) in South Georgia. We used a novel DSM-based method to facilitate automated and semi-automated counts of each colony to estimate population size. We assessed these DSM-derived counts against other popular counting and post-processing methodologies, including those from satellite imagery, and compared these to the results from four colonies counted manually to evaluate accuracy and effort. We randomly subsampled four colonies to test the most efficient and accurate methods for density-based counts, including at the colony edge, where population density is lower. Sub-sampling quadrats (each 25 m2) together with DSM-based counts offered the best compromise between accuracy and effort. Where high-resolution drone imagery was available, accuracy was within 3.5% of manual reference counts. DSM methods were more accurate than other established methods including estimation from satellite imagery and are applicable for population studies across other taxa worldwide. Results and methods will be used to inform and develop a long-term king penguin monitoring programme.