Procedure and architecture of the pipeline circuit for the calculation of the linear regression. Procedure and circuit architecture that allow to calculate the linear regression on a sequence of samples without having to store said samples previously in a memory. The calculation is carried out as the samples arrive at the system, thus achieving a low and constant latency from the instant in which the last sample arrives. On the other hand, the way to calculate the linear regression allows to obtain independently the parameters of the same: slope, ordered at the origin and error. In addition, thanks to the reuse of operations used, the number of operations is considerably reduced compared to the usual way of calculating the regression. (Machine-translation by Google Translate, not legally binding)