Automatic earthquake detection and phase picking from continuous waveform recorded by seismograms are highly important in seismology. Conventional methods contain short-term/long-term average (STA/LTA) and pattern recognition methods. However, recently, convolutional neural networks emerged as a game changer in machine learning as well as seismology, earthquake detection and phase picking. This study aims to evaluate these methods in a real case study. To this end, we implement two convolutional networks on 11 days continuous seismic waveforms of three seismometers recorded the Varzaghan-Ahar earthquake (11-August-2012), and detect seismic events and pick their corresponding seismic phases. Results show that relative to conventional methods (STA/LTA), the number of detected earthquakes are increased by 260% with a detection accuracy of more than 88%. Also, Pg and Sg phases are picked with an accuracy of more than 96 and 90%, respectively.