We are interesting in methods for predicting the time that a customer must wait at this arrival in multi-skilled call centers. We propose a free-parameters delay history predictor that can be used in multi-skilled contexts. It only uses the waiting times of previous customers in the same class who found the same length of queue when they joined the system, and computes a weighted average of this past delays. In our numerical experiments with simulation models and real multi-skill call center, this new predictor is very competitive with existing delay history predictors. It gives often better results than other delay history predictors, and it is also very easy to implement in practice. This delay predictors can also be used in other service systems such as medical clinic, bank or emergency service.