Severe slugging is one of the flow assurance concerns in oil and gas industry because of the undesired fluctuations of pressure and flow rates in the system. Knowledge of severe slugging frequency is essential for the receiving facility design. However, unlike the considerable investigations of normal slug frequency in horizontal and inclined pipes, little work has been done to predict severe slugging frequency. In this work, an experimental study of air-water two-phase flow was conducted in a long pipeline-riser system. According to the riser pressure drop signals and the power spectral density (PSD) distributions, flow patterns were classified into classical severe slugging, transitional severe slugging, oscillating flow and stable flow. The effect of riser-top choking on classical and transitional severe slugging frequency were analyzed. Based on a broad range experimental database from ten flow loops, the effect of dimensionless pipeline length on severe slugging frequency was analyzed and an empirical correlation was developed to predict severe slugging frequency. Good matches with literature data demonstrated the validation of the empirical correlation. The severe slugging frequency correlation was applied to predict severe slugging length and derive a new transition criterion between classical severe slugging and transitional severe slugging in the flow pattern map.