Disclosed is a method and an arrangement implementing the method for continuously estimating blood pressure. The method includes receiving an electrocardiography signal and a first photoplethysmography signal measured from the person, extracting values of a timing parameter set based on timings of the electrocardiography signal and the photoplethysmography signal, and calculating at least two intermediate estimates of the blood pressure of the person with linear regression models that model are based on different timing parameter subsets of the timing parameter set. At least one final estimate is then calculated on the basis of the intermediate estimates.