Various earthquake early warning (EEW) systems have been used in many countries. For an on‐site EEW method, earthquake magnitude and hypocentral distance are estimated from the first 3 s of P waves at the EEW site. Then S‐wave parameters, such as peak ground velocity (PGV), are usually computed from a ground‐motion prediction equation (GMPE) or a similar prediction model. Each step of the calculations brings in considerable uncertainties. In this study, we use the S‐ and P‐wave response spectral ratio as an EEW parameter. This study aims to develop an on‐site EEW model that will bypass the requirement of magnitude, distances, and earthquake tectonic locations under a set of conditions. Two empirical models are derived: one for the response spectral ratios between the first 3 s of S wave (after S‐wave arrival time) and the first 3 s of P wave ( model) and the other model is for the spectral ratios between the full record and the first 3 s of P‐wave ( model). The effects of earthquake focal mechanisms are substantially reduced at many spectral periods, and the event tectonic locations are not required at all spectral periods. Both models are reduced to a constant plus a site term when magnitude, depth, and distance are within a given set of ranges; the corresponding ranges of these parameters for the model are larger than those for the model. When magnitude and source distance are required, the effect of their errors on the EEW parameter is reduced significantly from those using a GMPE. Model provides a lower bound for model , and the two models predict very similar S‐ and P‐wave spectral ratios at short periods in the useful magnitude and distance ranges for practical EEW applications. Alarm issue rates for these two models are evaluated.