To　improve　the　human　gait　recognition　rate,　a　method　based　on　the　modified　gait　flow　image　(GFI)　considering　the　walking　direction　is　used　in　this　paper.　The　walking　direction　depends　on　the　height　and　position　at　the　walking　beginning　point　and　ending　point　in　a　gait　cycle.　GFI　is　generated　to　represent　the　motion　characteristics　of　a　gait　without　constructing　any　model.　Lacus-Kanade＇s　approach　is　used　to　calculate　the　optical　flow　field　of　GFI.　In　recognition　phase,　a　database　about　GFIs　at　different　walking　directions　of　the　target　is　established　in　advance.　After　computing　the　new　person＇s　walking　direction　and　GFI,　we　look　up　the　database　to　find　GFI　of　the　target　according　to　the　new　person＇s　walking　direction.　Then　Linear　Discriminant　Analysis　(LDA)　is　used　to　reduce　the　dimension.　At　last　we　can　recognize　whether　the　new　person　is　the　target　through　computing　the　similarity　between　the　new　GFI　and　the　target＇s　GFI.　The　performance　of　this　method　was　evaluated　on　the　data　in　the　CASIA　B　database　and　the　data　obtained　in　lab　environment.　The　experimental　results　prove　that　the　estimated　walking　direction　is　quite　correct,　and　the　GFI　computed　by　Lacus-Kanade＇s　approach　is　an　efficient　gait　representation　for　gait　recognition.　©　2012　IEEE.