Natural gas recovery from low permeability unconventional reservoirs - enabled by advanced horizontal drilling and multi-stage hydraulic fracture treatment - has become a very important energy resource in the past decade. While evaluating early gas production data in order to assess likely rate decline and ultimate gas recovery has been reported in literature, flowback water recovery has been given little consideration. Fracture fluid flowback is defined herein as aqueous phase produced within three weeks following a fracture treatment (exclusive of well shut-in time). Field data from Marcellus Shale wells in Northeastern West Virginia indicated about 2-26% of the fracture fluid is recovered during flowback. However, stimulation of gas shale is a complex engineered process, and the factors that control the volumetric flowback performance are not well understood. The objective of this paper is to use post-hoc analysis to identify correlations between fracture fluid flowback and attributes of well completion and geological setting, and to identify those factors most important in predicting flowback performances. To accomplish this objective we selected a representative subset of 187 wells for which complete data are available (from a full set of 631 wells), including well location, completion data, hydraulic fracture treatment data and production data. The wells were classified into four groups based on geological settings. For each geological group, engineering and statistical analyses were applied to study the correlation between flowback data and well completion through traditional regression methods. Important factors considered to affect flowback water recovery efficiency include number of hydraulic fracture stages, lateral length, vertical depth, proppant mass applied, proppant size, fracture fluid volume applied, treatment rate, and shut-in time. The total proppant mass, proppant size and shut-in time have relatively large influence on volumetric flowback performance. The new results enable one to estimate flowback volume in a spatial domain, based on known geological conditions and completion parameters, and lead to a better understanding of flowback behaviors in Marcellus Shale. This also helps industry manage flowback water and optimize production operations.