The regional phase Lg is often used to estimate location and magnitude for sources closer than 1,500 km. The complexity of Lg waveforms makes it difficult to determine consistently a unique Lg arrival time, thus affecting source location with a single station or array. This study tests an automatic method for timing Lg arrivals once the phase is roughly associated by analyst or using automatic detection algorithms. We used wavelet transforms to decompose the Lg signal into its components localized both in time and scale. A continuous wavelet transform (CWT) using a Daubechies order two (db2) wavelet is applied to 10 seconds of raw data containing the start of Lg. The coefficients at scale 16 from the db2 decomposition are squared and the resulting time series is represented by an approximation of the fourth-level discrete wavelet transform (DWT) using a Haar wavelet. A threshold detector is then applied to the resulting time series to determine the Lg arrival time. The method was tested using well located earthquakes and explosions from known mines (mb < 4.0), recorded on the vertical components at TXAR (Lajitas, Texas) and PDAR (Pinedale, Wyoming) arrays. The Lg arrival time was automatically picked with a standard deviation of less than 1.5 seconds (less than 10 km location error) for well known locations. Location standard deviation is larger with the increase of distance and smaller with the increase in signal to noise ratio of events.