As a paradigm of dynamic data-driven application systems (DDDAS) in nuclear power generation, this paper addresses the critical issue of on-line condition monitoring and early detection of anomalous behavior in boiling water reactor (BWR) nuclear power plants. The operation and maintenance in BWR plants are characterized by plethora of synergistic complex systems, where the energy is generated via a single heat removal loop and thus the real-time execution of condition monitoring and early detection of anomalous behavior become very challenging tasks. From the above perspectives, DDDAS methods have the potential of providing attractive solutions to the problems of on-line condition monitoring and anomaly detection. A symbolic data analysis (SDA) method has been adopted in this paper, where symbol sequences are generated by partitioning (finite-length) time series of plant data. Then, a special class of probabilistic finite state automata (PFSA), called D-Markov machine, is constructed to extract pertinent features from the statistics of time series as state probability vectors that serve as features for classification of different anomaly patterns. The proposed method is validated on a set of measurements taken from three different experiments on a BWR test facility, which emulate various types of loss-of-coolant-accidents (LOCA). Results exhibit the ability of SDA to extract, in a fast and accurate way, pertinent features for identifying the type of LOCA that may occur in a commercial-scale BWR.