Adaptive random testing (ART) ensures that test cases are more evenly distributed in the input domain, and thus achieves significantly stronger failure detection capability than the basic random testing.  Among the existing ART methods, the fixedsizecandidateset ART (FSCSART) exhibits better failure detection capability and has extensive applications. However, its failure detection effectiveness deteriorates significantly with the increase of input domain dimensions. To solve this high dimension problem, two types of distances are taken into account while choosing a test case from the candidate set: one is the distance from each candidate point to the already executed test cases; the other is the distance from individual candidate point to the center point. The comprehensive consideration of distances can reduce the priority of the candidate points at the edge of the input domain and overcome the disadvantage of the FSCSART. Experimental results show that the improved algorithm achieves a stronger failure detection capability in high dimensional input domains.