Dynamical models for 17 early-type galaxies in the Coma cluster are presented .
The galaxy sample consists of flattened , rotating as well as non-rotating early-types including cD and S0 galaxies with luminosities between M _ { B } = -18.79 and M _ { B } = -22.56 .
Kinematical long-slit observations cover at least the major and minor axis and extend to 1 - 4 r _ { \mathrm { eff } } .
Axisymmetric Schwarzschild models are used to derive stellar mass-to-light ratios and dark halo parameters .
In every galaxy the best fit with dark matter matches the data better than the best fit without .
The statistical significance is over 95 percent for 8 galaxies , around 90 percent for 5 galaxies and for four galaxies it is not significant .
For the highly significant cases systematic deviations between observed and modelled kinematics are clearly seen ; for the remaining galaxies differences are more statistical in nature .
Best-fit models contain 10-50 percent dark matter inside the half-light radius .
The central dark matter density is at least one order of magnitude lower than the luminous mass density , independent of the assumed dark matter density profile .
The central phase-space density of dark matter is often orders of magnitude lower than in the luminous component , especially when the halo core radius is large .
The orbital system of the stars along the major-axis is slightly dominated by radial motions .
Some galaxies show tangential anisotropy along the minor-axis , which is correlated with the minor-axis Gauss-Hermite coefficient H _ { 4 } .
Changing the balance between data-fit and regularisation constraints does not change the reconstructed mass structure significantly : model anisotropies tend to strengthen if the weight on regularisation is reduced , but the general property of a galaxy to be radially or tangentially anisotropic , respectively , does not change .
This paper is aimed to set the basis for a subsequent detailed analysis of luminous and dark matter scaling relations , orbital dynamics and stellar populations .