The COSMOS field has been the subject of a wide range of observations, with a number of studies focusing on reconstructing the 3D dark matter density field. Typically, these studies have focused on one given method or tracer. In this paper, we reconstruct the distribution of mass in the COSMOS field out to a redshift z= 1 by combining Hubble Space Telescope weak lensing measurements with zCOSMOS spectroscopic measurements of galaxy clustering. The distribution of galaxies traces the distribution of mass with high resolution (particularly in redshift, which is not possible with lensing), and the lensing data empirically calibrates the mass normalization (bypassing the need for theoretical models). Two steps are needed to convert a galaxy survey into a density field. The first step is to create a smooth field from the galaxy positions, which is a point field. We investigate four possible methods for this: (i) Gaussian smoothing, (ii) convolution with truncated isothermal sphere, (iii) fifth nearest neighbour smoothing and (iv) a multiscale entropy method. The second step is to rescale this density field using a bias prescription. We calculate the optimal bias scaling for each method by comparing predictions from the smoothed density field with the measured weak lensing data, on a galaxy-by-galaxy basis. In general, we find scale-independent bias for all the smoothing schemes, to a precision of 10 per cent. For the nearest neighbour smoothing case, we find the bias to be 2.51 ± 0.25. We also find evidence for a strongly evolving bias, increasing by a factor of ∼3.5 between redshifts 0 < z < 0.8. We believe this strong evolution can be explained by the fact that we use a flux limited sample to build the density field.