This study endeavors to extend current work on written error correction by investigating and comparing the effect of one traditional type of feedback—direct corrective feedback (DCF)—with an innovative type of error correction—feedback provided within context named situated feedback (SF) on collocation learning. The effects of the two feedback options were measured by a Translation Test and a Multiple Choice Test completed by 73 low-intermediate EFL students in China. The study found that both treatment groups performed much better than the control group in the posttests and delayed posttests and that there are significant advantages of SF group in comparison to DCF group in both posttests, suggesting that the provision of error correction within context is much more beneficial for learners in mastering target structures than direct correction alone and this superior effect can be retained for a long time. The implications of these findings from the perspective of cognitive psychology are discussed.