The purpose of this study was to investigate the effects of feedback type, problem presentation, and practice on problem-solving. One hundred and forty-seven college students were randomly assigned to 3 (feedback type) × 2 (problem presentation) experimental conditions to solve three problems for three times. The results showed that when word problems were presented with graphic representational cues, the time would decrease significantly and the accuracy increased. Repetitive practice alone did not improve the performance. In addition, only when practice was augmented with diagnostic feedback would the accuracy rate increase and efforts to solve the same problem repetitively be sustained, but such effects would be significant only after two rounds of practice. The effects of confirmatory feedback on problem-solving were similar to the control group in that it speeded up problem-solving, but there was little improvement in accuracy across practices. Analyses of 71 participants’ problem-solving protocols according to a two-space four-stage model showed that diagnostic feedback would direct participants’ cognitive activities to problem comprehension and problem definition, while confirmatory feedback would induce participants to put more efforts in computational analyses. This pattern of findings suggests that in solving ill-defined problems, numerical analysis is conducted only when the problem has been clearly defined.