The spread of social networking services has created an increasing demand for selecting, editing, and generating impressive images. This trend increases the importance of evaluating image aesthetics as a complementary function of automatic image processing. We propose a multi-patch method, named MPA-Net (Multi-Patch Aggregation Network), to predict image aesthetics scores by maintaining the original aspect ratios of contents in the images. Through an experiment involving the large-scale AVA dataset, which contains 250,000 images, we show that the effectiveness of the equal-interval multi-patch selection approach for aesthetics score prediction is significant compared to the single-patch prediction and random patch selection approaches. For this dataset, MPA-Net outperforms the neural image assessment algorithm, which was regarded as a baseline method. In particular, MPA-Net yields a 0.073 (11.5%) higher linear correlation coefficient (LCC) of aesthetics scores and a 0.088 (14.4%) higher Spearman's rank correlation coefficient (SRCC). MPA-Net also reduces the mean square error (MSE) by 0.0115 (4.18%) and achieves results for the LCC and SRCC that are comparable to those of the state-of-the-art continuous aesthetics score prediction methods. Most notably, MPA-Net yields a significant lower MSE especially for images with aspect ratios far from 1.0, indicating that MPA-Net is useful for a wide range of image aspect ratios. MPA-Net uses only images and does not require external information during the training nor prediction stages. Therefore, MPA-Net has great potential for applications aside from aesthetics score prediction such as other human subjectivity prediction.