This study examined whether visual-spatial attention could predict Chinese reading achievement longitudinally beyond some well-established reading precursors. A total of 257 second- and third-grade Hong Kong Chinese-speaking children participated in the study. Visual-spatial attention was measured by visual search tasks, which assessed the skills in highly inefficient and efficient searches. Chinese character reading, word reading fluency and reading comprehension skills were measured 1 year later. Latent variable path analysis showed that the skill in inefficient search uniquely predicted later word reading after the model controlled for phonological awareness, vocabulary knowledge, orthographic awareness, age and nonverbal IQ; while its effect on reading comprehension was completely mediated by word reading. Moreover, the skill in highly efficient search was not a unique predictor of both reading skills. The findings indicated the roles of top-down visual-spatial attention for word- and text-level Chinese reading.