In this paper we propose an output-bit selection technique for test response compaction, with which only a subset of output response bits is selected for observation during testing. Advantages of this technique include zero aliasing, high compaction ratio, full X-Tolerance, low area overhead, simple test control and high diagnosability. Also no circuit/ ATPG modification is needed, hence this work can be easily integrated into any typical industrial design/test flow to significantly reduce test cost. Experimental results show that in general less than 10% of test response data of already very compact test sets are needed to detect all testable stuck-At or transition faults, with the reduction ratio increasing with the size of circuits, e.g., only 1.27% of output bits need be observed for b19 that contains more than 1M faults. Efficient test architectures to implement this technique are also presented, which include one that can deal with test responses containing high percentage of unknown values.