Catastrophic forgetting is the notorious vulnerability of neural networks to the change of the data distribution while learning. This phenomena has long been considered a major obstacle for allowing the use of learning agents in realistic continual learning settings. Although this vulnerability of neural networks is widely investigated, it is currently only mitigated by explicitly reacting to the change of task. We suggest a novel approach for overcoming catastrophic forgetting in neural networks, using an online version of the variational Bayes method. Having a confidence measure of the weights alleviates catastrophic forgetting and, for the first time, succeeds in this even without the knowledge of when the tasks are being switched.