The mean wall shear stress, τ¯ w, is a fundamental variable for characterizing turbulent boundary layers. Ideally, τ¯ w is measured by a direct means and the use of floating elements has long been proposed. However, previous such devices have proven to be problematic due to low signal-to-noise ratios. In this paper, we present new direct measurements of τ¯ w where high signal-to-noise ratios are achieved using a new design of a large-scale floating element with a surface area of 3 m (streamwise) × 1 m (spanwise). These dimensions ensure a strong measurement signal, while any error associated with an integral measurement of τ¯ w is negligible in Melbourne’s large-scale turbulent boundary layer facility. Wall-drag induced by both smooth- and rough-wall zero-pressure-gradient flows are considered. Results for the smooth-wall friction coefficient, Cf≡ τ¯ w/ q∞, follow a Coles–Fernholz relation Cf=[1/κln(Reθ)+C]-2 to within 3 % (κ= 0.38 and C= 3.7) for a momentum thickness-based Reynolds number, Reθ> 15 , 000. The agreement improves for higher Reynolds numbers to <1 % deviation for Reθ> 38 , 000. This smooth-wall benchmark verification of the experimental apparatus is critical before attempting any rough-wall studies. For a rough-wall configuration with P36 grit sandpaper, measurements were performed for 10 , 500 < Reθ< 88 , 500 , for which the wall-drag indicates the anticipated trend from the transitionally to the fully rough regime.