Van den Broek's landscape model explicitly posits sequences of moves during reading in real time. Two other models that implicitly describe sequences of processes during reading are tested in the present research. Coded think-aloud data from 24 undergraduate students reading scientific text were analysed with lag-sequential techniques to compare specific transitions and flexibility with transitions between students who gained much in learning from reading a passage (n?=?11) and those who gained little (n?=?13). Just before verbalising inferences, those who gained much verbalised vocabulary, background knowledge, and strategies significantly more than did those who gained little. Those who gained much showed more flexible patterns before verbalising strategies and vocabulary, but rigid patterns just before inferences, whereas those who gained little showed the opposite pattern. Results both support and point out weaknesses in contemporary theories of reading comprehension and may explain some results from classroom experimental work.