In this study, new intensity prediction equations are derived for Central Asia, considering about 6000 intensity data points from 66 earthquakes encompassing the surface-wave magnitude range of 4.6-8.3. The suitability of the functional form used for constructing the model is assessed by comparing its predictions with those achieved through a non-parametric model. The parametric regressions are performed considering different measures of the source-to-site distance, namely the hypocentral, epicentral and the extended distance metrics. The latter is defined as the minimum distance from the site to a line crossing the epicentres, oriented along the strike of the earthquake and having a length estimated from the event's magnitude. Although the extended distance is introduced as a preliminary attempt to improve the prediction capability of the model by considering the finiteness of the fault extension, the standard deviation of the residual distribution obtained considering the extended distance (σ= 0.734) does not show an improvement with respect to the results for the epicentral distance (σ= 0.737). The similarity of the two models in term of average residuals is also confirmed by comparing the interevent errors obtained for the two regressions, obtaining very similar values for all earthquakes but the 1911, M 8.2 Kemin event. In particular, different evidences suggest that the magnitude of this event could be overestimated by about half a magnitude unit. Regarding the variability of the residual distribution, all the three considered components (i.e. interevent, interlocation and record-to-record variances) are not negligible, although the largest contribution is related to the record-to-record variability, suggesting that both source and propagation as well as site effects not captured by the considered model influence the spatial variability of the intensity values.