In the bioinformatics literature, pairwise sequence alignment methods appear with many variations and diverse applications. With this abundance, comes not only an emphasis on speed and memory efficiency, but also a need for assigning confidence to the computed alignments through p-value estimation, especially for important segment pairs within an alignment. This paper examines an empirical technique, called SEPA, for approximate p-value estimation based on statistically large number of observations over randomly generated sequences. Our empirical studies show that the technique remains effective in identifying biological correlations even in sequences of low similarities and large expected gaps, and the experimental results shown here point to many interesting insights and features.