Mapping land cover in complex regions like Western Cameroon Highlands is highly difficult. Most widely used algorithms are not easily implementable because of strong similarities observed in reflectance of different land cover units. This study evaluates the contribution of neural networks to the classification of LANDSAT 8 OLI images in order to achieve a better land cover map in this region. Image processing techniques (calculation of indices, principal components analysis, and color compositions) and a field survey allowed to discriminate and select trainings and validation sites of the main land cover units. Then, a network with 14 neurons in the input layer and 8 neurons in the output layer corresponding to different land cover classes was constructed. External and internal network parameters were experimentally selected for classification. The resulting map was finally validated with an overall accuracy of 90, 08% and a Kappa equal to 0.88. Eight land cover units have been identified. These are degraded forests, savannahs, bare soil and localities, water, wetlands, crops and burned areas. Finally, comparison with the maximum likelihood method has shown the superiority of neural networks with an overall accuracy difference of around 8%.