The estimate of the base flow index (BFI) based on the Hydrology of Soil Types (HOST) classification, BFIHOST, provides a measure of catchment responsiveness. BFIHOST is used with other variables to estimate the median annual maximum flood (QMED) in the UK standard Flood Estimation Handbook (FEH) statistical method and is also an explanatory variable in ReFH2, the FEH design hydrograph package. The current estimates of BFIHOST are derived from a restricted linear model, and a number of issues in the catchment dataset have been identified since the original work in 1995. The BFI calculated through base flow separation tends to be underestimated in clay-dominated catchments, and the calculation technique performs poorly in ephemeral catchments or those with missing data. The pragmatic bounding of BFI coefficients for permeable soils overlying aquifer outcrops is also problematic for small catchments. This paper investigates alternative regression methods to improve base flow estimates using the HOST class data for 991 stations (compared to 575 in the original); beta regression was found to give the best performance. Combining multiple rare classes into single classes is also shown to improve performance. The new version of BFIHOST was applied to the QMED equation, showing improved performance.