We have observed that the collisional frequency shift in primary caesium fountain clocks varies with the clock state population composition and, in particular, is zero for a given fraction of the |F = 4, m F = 0〉 atoms, depending on the initial cloud parameters. We present a theoretical model explaining our observations. The possibility of the collisional shift cancellation implies an improvement in the performance of caesium fountain standards and a simplification in their operation. Our results also have implications for test operation of fountains at multiple π/2 pulse areas. The primary caesium clocks have increased their accuracy and stability by an order of magnitude over the last decade [1]. This staggering progress has mainly resulted from the implementations of laser cooling and the fountain configuration. Slow atoms allow for long interrogation times and observation of narrow resonances (≤ 1 Hz) leading to improvements in the short-term stability. At the same time many systematic effects, which limited the performance of thermal beam clocks, have been significantly reduced in the fountains resulting in improvements of their accuracy. The use of ultracold atoms gave rise, however, to a frequency shift due to collisions, an effect generally neglected in thermal beam clocks. For the several operational fountain frequency standards, the collisions constitute one of the major systematic effects limiting the standards' performance. Generally, the fountain standards are corrected for the collisional frequency shift by extrapolating the measured frequency to zero atom density, assuming a linear dependence between the shift and the density. However, the atomic density is not measured directly in the fountains. The changes in the density are derived from changes in the number N at of detected atoms. A potential deviation from linearity between N at and the atomic density (and hence the shift) gives rise to an uncertainty of the correction. One way to minimize this uncertainty is to operate the fountain at a very low density [2]. Unfortunately, with a low atom number the detection signal-to-noise ratio (and short-term stability of the standard) is reduced. Another way relies on the implementation of a technique based on adiabatic passage to link unambiguously changes in density with changes in N at [3]. In this case, the accuracy of the collisional shift measurement is ultimately limited by a residual amount of atoms in |F = 3, m F ≠ 0〉 colliding with atoms in the clock states during the ballistic flight [4]. Ways of cancelling the collisional shift have been studied earlier. The cancellation would be possible if the cross-sections of the frequency changing collisions of the two clock states |3,0〉 and |4,0〉 were of opposite sign. Then, if a certain composition of the clock states were excited the two contributions to the clock shift would cancel. In the 1990s the relevant cross-sections for various Cs isotopes were calculated and it was shown that the