Velocity-related problems like Doppler shifts, distributed cavity phase-shift and pulling by neighbouring transitions are presently the limiting factors in the accuracy of Cs beam primary frequency standards. The use of slow atoms appears to be the only fundamental solution to these problems, and new kinds of standards are being investigated, based on different ideas for trapping and cooling down the atoms. The thirty-year-old idea of gravitational trapping was abandoned after a short trial due to bad signal-to-noise ratio in a Cs fountain beam which made use of magnetic state selection and hot wire detection. In this paper it is shown that optical state selection and detection, in conjunction with an open Fabry-Perot type resonator, can make that old idea very attractive again. A rough design sketch is outlined and a detailed analysis is given of the various uncertainty sources affecting the frequency of a fountain Cs standard, while problems which need further examination are pointed out.