This study aims at introducing the potential to utilise transfer learning methods in the training of artificial neural networks for tribological applications. Artificially enhanced surfaces through surface texturing, as an example, are investigated under hydrodynamic regime of lubrication. The performance of these surface features is assessed in terms of load carrying capacity and friction. A large performance dataset including bearing load carrying capacity and friction is initially obtained for a specific category of textures with rectangular cross-sectional profile through analytical methods. The produced bearing performance are used to train a neural network. This neural network was then trained further by a minimal set of performance measure data from an intended category of textures with triangular cross-sectional profiles. It is shown that the resulting neural network performs with acceptable level of confidence for those intended texture profiles when trained with such relatively low number of performance data points. The results indicate that fast analytical methods can potentially produce a large volume of training datasets, which effectively allows for use of relatively lower number of training data sets from the intended category, where creating data for trainings can be more complex or time consuming. Use of transfer learning method in tribological applications and use of bearing performance parameters, as opposed to bearing design parameters, for training the neural networks are the major novel contributions of this study, which has not hitherto been reported elsewhere.