Feature recognition is important for describing shapes in many applications taking advantage of solid modeling. Graph-based feature recognition methods search from solid models the unique patterns of features that are represented as a graph. A typical example of such patterns is a loop of concave edges. When the loop is an inner loop on a single face, it is a strong hint of the existence of protrusion feature and recognition of protrusion faces is straightforward. However, when a protrusion feature lies on multiple faces, it is bounded by a loop of concave edges that are not on a single face. Consequently, the rule of inner loop is no more available and recognition of protrusion faces becomes unclear. To address this problem, a new quantitative measure, orthogonal bounding factor (OBF), is introduced. OBF is defined as the sum of cross products of two consecutive vectors normal to a set of faces, and it physically represents the possibility of being a protrusion in a solid model. The formal definition of orthogonal bounding factor is established and a method to recognize protrusion features using OBF is presented. Examples are also shown to demonstrate the effectiveness of the method for feature recognition.