Determination of power quality becomes more and more important in the future. Low voltage networks are usually large and very complex. Therefore the calculation of power quality parameters by modeling as equivalent network is hardly possible in practice. That's why new methods for efficient and exact estimation of power quality parameters in low voltage networks are necessary. The presented method is based on the fact that networks of similar structure have a similar behavior in power quality. The Points of Common Coupling (PCC) are divided into different classes, where each class consists of PCC's with similar characteristics. This way the method allows the estimation of power quality levels based on the class a PCC is assigned to. The paper demonstrates the method for the 5th voltage harmonic as an example power quality parameter. The above-mentioned classification of substations is based on probabilistic neural networks.