Given　the　rapidly　increasing　number　of　available　Web　services　and　the　different　user　QoS　requirement,　selecting　a　suitable　service　for　each　user　adaptively　is　difficulty　in　AmI　environment.　In　this　paper,　we　propose　a　backward　inference　based　user　QoS　requirements　self-learning　model　to　learn　the　users＇　special　preferences　by　using　the　service　evaluation　information　and　then　construct　a　user＇s　QoS　cognitive　set　by　means　of　learned　information.　We　also　put　forward　a　dynamic　adaptive　service　selection　algorithm　based　on　the　dynamic　multi-attribute　decision　making　theory.　The　algorithm,　firstly,　can　generate　user＇s　QoS　requirement　preferences　according　to　the　information　of　candidate　services　and　user＇s　QoS　cognitive　set.　Then　it　automatic　selects　a　suitable　service　for　user　by　using　the　services＇　history　information,　service　providers＇　statements　values　and　user　QoS　requirements.　Last,　we　verify　the　usability　of　the　model　through　prototypical　implementation　and　explain　its　advantage　and　deficiency.　©　2010　Binary　Information　Press.