•The use of low-noise surfaces is the best solution to mitigate tyre/road noise.•Green Public Procurement (GPP) requires CPX measurements on new low-noise pavements.•Modelling CPX levels in function of the mix properties is necessary to reduce costs.•Two models were elaborated, using two frequency separations for tyre/road noise.•The variables used are related to volumetric and composition properties of the mix. Road traffic in urban contexts produces noise mainly by the interaction of tyres with pavement surface and, therefore, the use of low-noise surfaces represents the best solution since they aim to mitigate the source. Moreover, in urban contexts it is often the only viable solution, together with a careful traffic planning. The main challenge in their adoption as noise mitigation actions is to be able to forecast the acoustical performances that the new road surface will be able to offer. In the UE, the new Green Public Procurement requires experimental verification of noise performance compliance: the designer must declare the acoustical performance of the proposed low-noise pavement and, a few months after the laying, the actual performance of the road surface must be tested using the Close Proximity Method (CPX). Due to the importance of being able to forecast CPX levels, the present work reports a novel way to model CPX broadband levels of newly laid low-noise road surfaces using only data available to the designer before the laying or easily obtained through coring tests, such as grading curve, fractal dimension, asphalt binder content, air voids, voids in mineral aggregates. Two models were elaborated, using two different frequency separations for tyre/road noise. The first model separates low and high frequency contributions, while the second one also considers noise around 1 kHz separately, using a three-band model. Both models are capable of forecasting the acoustic performance of newly laid low-noise road surfaces, using different road mixture parameters at different frequency ranges. The three band model shows a lower RMSE.