Background: Mathematical word-problem solving involves both text comprehension and computation, but existing process models differ in their assumptions about whether these processes occur linearly or flexibly, and whether they are situated in integrated or separable mental representations. Aims: Using eye-tracking, we investigated how the mathematical and linguistic problem difficulty and abilities of the solver and the position of the question affected the solution process, using three categories of effects that can be explained by some models, but not the others. Sample: Participants were 50 undergraduate students. Methods: Participants solved 16 three-line traditional mathematical word problems that were independently varied by numerical, lexical, and syntactic problem difficulty and by the position of the question in a within-subject design. Abilities were assessed with standardized instruments. By using eye tracking, visual attention on text and numbers was distinguished in two solution phases and analyzed using linear mixed models. Results: Linguistic problem difficulty and abilities affected attention on text in the first solution phase, and mathematical problem difficulty and abilities affected attention on numbers and text in the second solution phase. Putting the question first lead to less attention on numbers and text during the first solution phase, but more attention in the second phase. Conclusions: Our results concur with models that assume text comprehension precedes computation but occurs in a shared mental representation. However, findings regarding the question position indicate that some mathematical processes occur already in the first solution phase.