Autocorrelation (AC) is one of the most frequently used methods for initial acquisition. Usually, algorithms that employ a metric based on AC for this purpose, focus on the estimation of synchronization parameters, ignoring the detection performance. This paper analyses the relationship between the structure of the reference sequence that serves to compute the AC and the corresponding attainable detection performance. The distributions of test statistics based on AC for frame detection are derived and validated through simulations in a frequency-flat fading channel. It is shown that the use of AC can outperform, in terms of detection performance, the more complex matched filtering in a certain signal-to-noise ratio region, at the cost of significantly longer required sequences. Besides, we show that a set of sequences, with different periodic structures, can provide the same target detection probability, allowing a flexible sequence design.