Applications of soft-sensing techniques in modern complex process systems can significantly reduce the monitoring cost and minimize the chance of false alarm. However, the potential of soft-sensing techniques have yet to be adequately exploited in the realm of dynamic risk assessment. This paper attempts to close such a gap by proposing a novel synthesis of soft-sensing and loss function for dynamic risk assessment of complex process operations. The soft-sensing technique adopted in this work is based on a Bayesian formulation of the kernel regression which takes into account both the model uncertainty and data uncertainty. Such a probabilistic regression model is able to predict the probability of an undesired event in real-time. On the other hand, the potential economical losses associated with the undesired event is also quantified through loss function. Subsequently, the dynamic risk of an undesired event can be determined as the product of its occurrence probability and its associated potential economical losses. The effectiveness of the proposed dynamic risk assessment is demonstrated through an industrial case study. (C) 2016 Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.