We present an improved version of the POTENT method for reconstructing the
cosmological velocity and mass density fields from radial peculiar velociti
es, test it with mock catalogs, and apply it to the Mark III Catalog of Gal
axy Peculiar Velocities. The method is improved in several ways: (1) the in
homogeneous Malmquist bias is reduced by grouping and corrected statistical
ly in either forward or inverse analyses of inferred distances, (2) the smo
othing into a radial velocity held is optimized such that window and sampli
ng biases are reduced, (3) the density field is derived from the velocity h
eld using an improved weakly nonlinear approximation in Eulerian space, and
(4) the computational errors are made negligible compared to the other err
ors. The method is carefully tested and optimized using realistic mock cata
logs based on an N-body simulation that mimics our cosmological neighborhoo
d, and the remaining systematic and random errors are evaluated quantitativ
ely. The Mark III catalog, with similar to 3300 grouped galaxies, allows a
reliable reconstruction with fixed Gaussian smoothing of 10-12 h(-1) Mpc ou
t to similar to 60 h(-1) Mpc and beyond in some directions. We present maps
of the three-dimensional velocity and mass-density fields and the correspo
nding errors. The typical systematic and random errors in the density fluct
uations inside 40 h(-1) Mpc are +/-0.13 and +/-0.18 (for Omega = 1). In its
gross features, the recovered mass distribution resembles the galaxy distr
ibution in redshift surveys and the mass distribution in a similar POTENT a
nalysis of a complementary velocity catalog (SFI), including such features
as the Great Attractor, Perseus-Pisces, and the large void in between. The
reconstruction inside similar to 40 h(-1) Mpc is not affected much by a rev
ised calibration of the distance indicators (VM2, tailored to match the vel
ocities from the IRAS 1.2 Jy redshift survey). The volume-weighted bulk vel
ocity within the sphere of radius 50 h(-1) Mpc about the Local Group is V-5
0 = 370 +/- 110 km s(-1) (including systematic errors) and is shown to be m
ostly generated by external mass fluctuations. With the VM2 calibration, V-
50 is in a similar direction and reduced to 305 +/- 110 km s(-1).