We report results of two studies examining the relation between decoding and reading comprehension. Based on our analysis of prominent reading theories such as the Simple View of Reading (Gough & Tunmer, 1986), the Lexical Quality Hypothesis (Perfetti & Hart, 2002) and the Self-teaching Hypothesis (Share, 1995), we propose the Decoding Threshold Hypothesis, which posits that the relation between decoding and reading comprehension can only be reliably observed above a certain decoding threshold. In Study 1, the Decoding Threshold Hypothesis was tested in a sample of over 10,000 Grade 5-10 students. Using quantile regression, classification analysis (Receiver Operating Characteristics) and broken-line regression, we found a reliable decoding threshold value below which there was no relation between decoding and reading comprehension, and above which the two measures showed a positive linear relation. Study 2 is a longitudinal analysis of over 30,000 students' reading comprehension growth as a function of their initial decoding status. Results showed that scoring below the decoding threshold was associated with stagnant growth in reading comprehension. We argue that the Decoding Threshold Hypothesis has the potential to explain differences in the prominent reading theories in terms of the role of decoding in reading comprehension in students at Grade 5 and above. Furthermore, the identification of decoding threshold also has implications for reading practice. [This is the in press version of an article to be published in "Journal of Educational Psychology."]