Fifty-one severely to profoundly deaf students (mean dB hearing loss = 89) were randomly assigned to two groups that differed by the type of probes (short or long) used to elicit prior knowledge (PK). PK scores were used to predict reading comprehension (RC), which was assessed by students' responses to three types of questions: test-explicit (TE), text-implicit (TI), and script-implicit (SI). Multiple regression models with PK scores and scores from a standardized achievement test (Stanford Achievement Test - Hearing Impaired Version, reading subtest) were also used to predict RC. The regression model showed that, for the group pretested with an in-depth, or long, probe of PK, the best predictor of RC was the ability to answer TE and SI questions. We present discussions of the observed differences in comprehension as a function of long and short knowledge probes and the use of three question types, together with implications for instruction.