Owing　to　the　uncertainty　of　the　traffic　system　and　incomplete　travel　information,　travelers　usually　make　route-choice　decisions　relying　on　their　own　experience.　In　this　paper,　we　assume　that　commuters　make　their　route-choice　decisions　based　on　the　perceived　cost　in　a　logit-based　manner,　and　different　memory-based　learning　strategies　on　previous　travel　time,　such　as　smoothed　adaptive　pattern　and　peak-end　adaptive　pattern　(anchoring　on　highest　travel　time　or　lowest　travel　time),　are　proposed　to　obtain　the　perceived　cost.　A　numerical　example　is　also　given　for　comparing　the　impact　of　different　learning　strategies　on　flow　evolution　and　further　illustrating　the　model.　The　results　show　that　the　peak-end　adaptive　pattern　could　capture　the　commuters＇　risk　attitude　in　the　route　choice　process　and　thus　provide　a　more　actual　traffic　flow,　which　is　obviously　helpful　to　traffic　control.　©　2016　ASCE.