Abstract: Burst detection plays a crucial role for burst-mode demodulation, which helps the acquisition and synchronization of data bursts. Most conventional Data-Aided (DA) burst detection approaches demands timing, and the performance degrades severely in the presence of frequency offset. This paper proposes a burst detection algorithm based on the Average-Likelihood-Ratio-Test (ALRT) for over-sampled modulated signals with timing uncompleted. The proposed algorithm outperforms the previously proposed algorithms and is robust to the frequency offset. Furthermore, the dependence of false alarm rate on SNR is slight.