We investigate the clustering properties of loose groups in the Perseus--Pisces redshift Survey (PPS). Previous analyses based on CfA and SSRS surveys led to apparently contradictory results. We investigate the source of such discrepancies, finding satisfactory explanations for them. Furthermore, we find a definite signal of group clustering, whose amplitude $A_G$ exceeds the amplitude $A_g$ of galaxy clustering ($A_G=14.5^{+3.8}_{-3.0}$, $A_g=7.42^{+0.20}_{-0.19}$ for the most significant case; distances are measured in $\hMpc$). Groups are identified with the adaptive Friends--Of--Friends (FOF) algorithms HG (Huchra \& Geller 1982) and NW (Nolthenius \& White 1987), systematically varying all search parameters. Correlation strenght is especially sensitive to the sky--link $D_L$ (increasing for stricter normalization $D_0$), and to the (depth $\mlim$ of the) galaxy data. It is only moderately dependent on the galaxy luminosity function $\phi(L)$, while it is almost insensitive to the redshift--link $V_L$ (both to the normalization $V_0$ and to the scaling recipes HG or NW).