Within the framework of the theory of visual attention (TVA), the visual attention span (VAS) deficit among individuals with developmental dyslexia has been ascribed to the problems entailed by bottom-up (BotU) and top-down (TopD) attentional processes. The former involves two VAS subcomponents: the visual short-term memory storage and perceptual processing speed; the latter consists of the spatial bias of attentional weight and the inhibitory control. Then, what about the influences of the BotU and TopD components on reading? Are there differences in the roles of the two types of attentional processes in reading? This study addresses these issues by using two types of training tasks separately, corresponding to the BotU and TopD attentional components. Three groups of Chinese children with dyslexia--15 children each in the BotU training, TopD training, and non-trained active control groups were recruited here. Participants completed reading measures and a CombiTVA task which was used to estimate VAS subcomponents, before and after the training procedure. Results showed that BotU training improved both the within-category and between-category VAS subcomponents and sentence reading performance; meanwhile, TopD training enhanced character reading fluency through improving spatial attention capacity. Moreover, benefits on attentional capacities and reading skills in the two training groups were generally maintained three months after the intervention. The present findings revealed diverse patterns in the influences of VAS on reading within the TVA framework, which contributes to enriching the understanding of VAS-reading relation.