It　is　difficult　to　interpret　ultrasonic　signals　when　the　echoes　that　are　reflected　from　the　various　target　features　overlap.　In　this　paper,　a　pulse　compression　technique　based　on　a　combination　of　Wiener　filtering　and　autoregressive　(AR)　spectral　extrapolation　is　used　to　process　ultrasonic　signals　to　improve　their　temporal　resolution.　The　influence　of　the　AR　order　and　the　width　of　the　frequency　window　in　AR　spectral　extrapolation　are　discussed　in　detail,　and　optimal　parameters　are　obtained.　The　technique　is　then　applied　to　measure　the　size　of　the　defects　in　a　plate　and　discern　two　adjacent　defects　in　a　block　using　a　phased　array.　It　is　shown　that　the　widths　of　the　defects　in　the　plate　were　measured　accurately　and　the　two　adjacent　defects　were　distinguished　well　by　the　proposed　technique.