With the rapidly growing number of Web services, how to identify high quality Web services has become a hot topic. Quality of Service (QoS) is a key criterion for the choice of optimal Web services from a set of candidate Web services with similar functions. However, QoS data is acquired through the invocation of services from users. Thus, QoS prediction is critical for building high-quality service-oriented applications. Since QoS is highly related to dynamic factors such as users' or services' status and network environments which are variable over time, it is an important task to predict the unknown QoS values at runtime. In addition, the factors which cause the change of QoS may be various, such as the influence of noise, the change of the network environments. Prediction without taking account of these factors will affect the prediction accuracy. To address the problems above, we propose an online prediction approach for dynamic QoS (OPA-DQ). OPA-DQ extends matrix factorization into an online approach to make the QoS prediction process more efficient. According to the analysis on the factors which will cause the change of QoS, we build a series of processes to make better QoS prediction performance. Experimental results in a real world dataset indicate that our online approach has higher prediction accuracy and efficiency compared with other approaches.