As most mature earthquake faults are filled with gouge at shallow part and it’s difficult to accumulate pressure，continental earthquakes hardly occurred shallower than 5km in depth． However，several very shallow earthquakes at a depth of about 3km occurred in Sichuan and Chongqing in recent years，which might be related to the formation of new faults or industrial activities． Generally speaking，to locate focal depth accurately using arrival time of direct body waves，stations within the epicentral distance of two times the focal depth are needed． Thus，by now the density of station is not enough to locate the focal depth of very shallow earthquakes ( shallower than 5km) ． Under this circumstance，utilizing the earthquake waveform is very useful to constrain the focal depth． CAP (Cut and Paste) is a method widely used to invert earthquake focal mechanism，magnitude and depth using earthquake waveform． Although this method is not strict with accuracy of crustal velocity structure model，large deviation of the velocity structure model can lead to bad focal depth inversion result． We took the Ｒongchang earthquake as an example，which happened on September 10，2010 with Mb4． 7，to test the accuracy of focal depth inversion using a series of velocity structure models，with the teleseismic depth phases as a constraint． We found that the deviation of focal depth inversion is small if the deviation of velocity structure model is within 10% ． Nevertheless，the deviation is large if the shallow part of velocity structure model has a deviation of about 50% ． Thus，it is necessary to ensure the accuracy of shallow part of velocity structure model to locate focal depth of a very shallow earthquake．