Structural health monitoring is concerned with estimating the current health state of a structure being monitored and aims to provide reliable information on the presence, location, and severity of damage. When the structure experiences damage, it causes changes in structural parameters such as stiffness reduction and consequently alters measured signals or features extracted from the measured signals. Therefore, damage diagnosis can often be performed by novelty detection, i.e., detecting the changes in the measured signals or the features by comparing the most recent data obtained from an unknown condition of the structure with the baseline data accumulated from its normal conditions. In reality, time-varying environmental and operational conditions such as temperature, wind, and traffic loading also induce changes in the measured signals or the features and consequently may produce false alarms. Therefore, to achieve successful novelty detection, it is necessary to distinguish the signal changes caused by abnormality from those caused by environmental and operational variations. This process is called data normalization. In this study, kernel principal component analysis is employed to perform data normalization and incorporated with a novelty index and generalized extreme value statistics for novelty detection. The proposed approach is applied to the field data obtained from the Yeongjong grand bridge in Korea and demonstrated to be a promising tool for detecting abnormality in the presence of environmental and operational variations.