Firstly, the author generalizes the indiscernible relation to similarity relation, gives a definition of lambda-discernibility matrix; secondly, an attribute reduct algorithm of decision table with continuous condition attributes is put forward; thirdly, an algorithm of computing significance of each condition attribute is put forward according to the properties of lambda-discernibility matrix; at last, the time complexity of the attribute reduct algorithm is analyzed and the rationality and effectiveness of this algorithm is accounted for through an example.