Image memorability prediction involves two problems, feature representation and prediction model. Most of previous researches just focused on addressing the first problem by investigating the factors making an image memorable, and conducted feature fusion and regression learning in two separate steps. Results of feature fusion decide the performance of regression. Lack of using an integrated learning mechanism cannot efficiently address image memorability prediction tasks, since it may lead to sub-optimal prediction results. To solve the problem presented above, we introduce a novel image memorability prediction model based on low-rank representation learning. We seek the lowest-rank representation among all the samples by projecting the original feature matrix into a subspace spanned by a low-rank projection matrix. Meanwhile, we learn a regression coefficient to build connections between latent low-rank representations and memorability scores by linear regression. Furthermore, we develop an effective algorithm based on the augmented Lagrange multiplier method to solve our model. Extensive experiments conducted on publicly available image memorability datasets demonstrate the effectiveness of the proposed schemes.