Early intervention for children with reading impairments is crucial in order to achieve reading improvements and avoid school failure. One line of reading intervention research focuses on the experimental manipulation of reading rate through a text-fading training approach. Considering relevant reading-related predictors (i.e., orthographic knowledge and rapid automatized naming; RAN), we aim at evaluating the text-fading training's efficiency for a sample of German reading-impaired third graders (n = 120). The purpose of the present study was to examine (1) the predictive value of orthographic knowledge and RAN and their contribution of explained variance in comprehension performance during training, (2) text-fading training effects on reading rate and comprehension in a pre-post comparison, and (3) (lasting) text-fading training effects at word and sentence level in a pre-post-follow-up design. Results of structural models indicated RAN to be significantly related to comprehension performance for the experimental group, whereas no sufficient regression weight was found for orthographic knowledge. A reverse pattern was found for the self-paced group. No significant improvements regarding reading rate and comprehension were revealed for the experimental group after training. However, significant positive effects on word and sentence level at post-test time point indicate stronger reading improvements for the experimental compared to the control group. The retention of training gains was indicated at sentence-level reading 6 months after the training. Possible explanations for the presented positive training effects as well as the mixed results for reading rate, comprehension, and follow-up preservation are discussed.