The Southern Ocean (SO) is one of the cloudiest places on Earth, with distinct cloud properties including a high prevalence of multilayer clouds. Previous research has found that multilayer clouds contribute to net cloud radiative effect biases. In this paper we have compared and validated different LEO passive sensor retrievals (AVHRR-Patmos-X, CMSAF, MODIS collection 6.1) over the SO. We found that passive sensors can have both positive and negative biases in cloud top height (CTH) when compared to active sensors (± 2 km). One of the significant factors for the observed differences is multilayer clouds. We found CTHs identified as multilayer had over 3 times the bias compared to those identified as single layer. Given the results of the comparison and a need for more accurate cloud retrieval for multilayer clouds in particular, we developed a new multilayer retrieval algorithm for CTH from MODIS data over the SO region using an artificial neural network (ANN) approach. The retrieval algorithm employs MODIS radiances and reanalysis datasets. The algorithm’s performance for the topmost cloud layer demonstrates a significant improvement compared to the traditional retrieval approaches. The MODIS CTHs mean bias error against the CloudSAT- CALIOP merged dataset was reduced to approximately 20 m with an RMSE of 1 km. In case of multilayer scenarios, the CTHs of the lower layers were also retrieved for many of the multilayer scenes.