An　assembly　line　system　is　a　manufacturing　process　in　which　parts　are　added　in　sequence　from　workstation　to　workstation　until　the　final　assembly　is　produced.　In　a　mixed-model　assembly　line　balancing　problem,　tasks　belonging　to　different　product　models,　are　allocated　to　workstations　according　to　their　processing　times　and　precedence　relationships　amongst　tasks.　The　research　parlayed　two　features,　learning　effect　and　uncertain　demand,　into　the　conventional　mixed-model　assembly　line　balancing　model,　which　is　the　main　contribution　of　our　paper.　Both　features　can　affect　the　new　decision　appeared　in　the　stated　problem　-　the　level　of　production.　The　problem　setup　as　well　as　the　new　decisions　considered　in　the　problem　are　novel.　The　proposed　model　optimized　two　objectives,　total　expected　cost　and　average　cycle　time.　To　solve　the　model,　a　mixed　integer-based　heuristic　and　a　customized　variable　neighborhood　search　method　is　proposed.　The　algorithms　are　examined　for　two　different　system　response　time　requirements.　Computational　results　showed　that　the　mixed　integer-based　heuristic　is　more　efficient　if　there　is　enough　response　time　for　the　decision　making　process.　On　the　contrary,　the　customized　variable　neighborhood　search　method　can　deliver　promising　results　under　real-time　conditions.　The　Pareto-optimal　set　can　be　generated,　which　provides　the　managers　with　multiple　choices　for　different　cost　and　cycle　time　combinations.