Applying the blind detection of transmitted symbols, the receiver can enhance the communication throughput, since the blind detection does not need pilot symbols. For the orthogonal frequency-division multiplexing (OFDM) system, which is widely used in modern wireless communications, some blind channel response (CR) estimation algorithms have been proposed. There are also blind detection algorithms that directly detect the transmitted data without explicit CR estimation. However, not all blind algorithms can be applied in time-varying or frequency-selective fading channels. Moreover, the blind algorithms usually have higher complexity. In this paper, we first propose a general principle, by which one can build a metric for blind detection from any pilot-assisted CR estimation algorithm. By this principle and based on the least-squares-fitting (LSF) pilot-assisted CR estimation, we derive a metric for blind detection and propose an efficient blind data detection algorithm. The proposed algorithm is operated on a block of received symbols. The block can be composed of symbols along subchannels or along time slots, and the blind detection can be applied in frequency-selective or time-varying fading channels. With the proposed tree search method, the proposed blind detection algorithm attains the same performance as a previous blind detection algorithm based on the same metric, while the complexity is significantly reduced.