Amid predictors of second language (L2) reading, both first language (L1) reading and L2 listening embody the complexities of comprehension skill in their construct. Using two layers of latent-trait assays, this inquiry explored L1 reading and L2 listening together with L2 knowledge, which is often displayed to strongly forecast L2 reading. Data from 325 L1-Japanese university students were first Rasch-analysed to attain multiple latent indicators of criterion and predictor variables, which were then submitted to Structural Equation Modelling to assess the relative weights of the latent predictors. Over 90 per cent of the L2 reading variance was jointly accounted for by the predictors, with L2 listening and L2 knowledge explaining most of the variance. A separate assay of the higher proficiency subsample ended in L2 knowledge displaying the largest contribution, followed by L2 listening and finally by L1 reading.