The effective detection and classification of damage in complex structures is an important task in the realization of structural health monitoring (SHM) systems. Conventional information processing techniques utilize statistical modeling machinery that requires large amounts of 'training' data which is usually difficult to obtain, leading to compromised system performance under these data-scarce conditions. However, in many SHM scenarios a modest amount of data may be available from a few different but related experiments. In this paper, a new structural damage classification method is proposed that makes use of statistics from related task(s) to improve the classification performance on a data set with limited training examples. The approach is based on the framework of transfer learning (TL) which provides a mechanism for information transfer between related learning tasks. The utility of the proposed method is demonstrated for the classification of fatigue damage in an aluminum lug joint.