In　this　paper,　we　present　an　investigation　on　heuristics　retrieval　in　human　problem　solving　by　combining　the　computational　cognitive　model　ACT-R　(Adaptive　Control　of　Thought-Rational)　and　advanced　fMRI　(functional　Magnetic　Resonance　Imaging)　brain　imaging　technique.　As　a　new　paradigm,　4*4　Sudoku　is　developed　to　facilitate　this　study,　in　which　seven　heuristics　that　can　be　classified　into　3　groups　are　designed　to　solve　two　types　of　tasks:　simple　and　complex　ones.　The　cognitive　processes　of　the　two　types　of　4*4　Sudoku　tasks　are　explored　based　on　the　outputs　of　ACT-R　model.　This　study　shows　that　several　key　elements　take　important　roles　in　the　retrieval　of　heuristics,　including　the　ways　of　problem　presentation,　complexity　of　heuristics　and　status　of　goal.　The　fitness　of　model　prediction　to　real　participants＇　data　on　behavior　and　BOLD　(Blood　Oxygenation　Level-Dependent)　response　in　five　predefined　brain　regions　illustrates　that　our　hypotheses　and　results　are　acceptable.　This　work　is　a　significant　step　towards　tackling　the　puzzle　of　the　heuristics　retrieval　in　human　brain.　©　2009　Springer-Verlag　Berlin　Heidelberg.