The main source of information on the abundance of polymetallic nodules (APN) is the results of direct seafloor sampling, mainly using box corers. Due to the vast spread of nodule occurrence in the Pacific, the distances between successive sampling sites are significant. This makes it difficult to reliably estimate the nodule resources, especially in parts of the deposit with small areas corresponding to the areas scheduled for extraction in the short term (e.g. within one year). It seems justified to try to increase the accuracy of nodule resource estimates through the use of information provided by numerous photos of the ocean floor taken between sampling stations. In particular, the percentage of nodule coverage of the ocean floor (NC), the data on fraction distribution of nodules (FD) and the coverage of nodules with sediments (SC) are important here. In the presented study, three regression models were used to predict the nodule abundance from images: simple linear regression (SLR), multiple regression (MR), and general linear model (GLM). The GLM provides the most accurate prediction of nodule abundance (APN) due to the ability of this model to simultaneously take into account both quantitative variable (NC) and qualitative variables (FD, SC). The mean absolute errors of APN prediction are in the range of 1.0–1.7 kg/m2, which is 7–13% of the average nodule abundance determined for training or testing data sets. This result can be considered satisfactory for predicting the abundance in ocean floor areas covered only by photographic survey.