In order to gain insight into droplet behavior at the edge of clouds, a laboratory experiment has been carried out to study the conditioned statistics of inertial sub-Kolmogorov particles in shearless turbulent–non-turbulent and turbulent–turbulent mixing layers. The water droplets were injected from the homogeneous turbulence side of the flow, and their velocity and size distribution profiles were measured by a combined LDV/PDPA technique. The fluid velocity field was measured using hot-wire anemometry in the droplet-free flow. A conditioning method with the stream-wise velocity chosen as a turbulence detector function was used to identify the turbulent regions in the mixing layer. The particle concentration profiles, mass fluxes and small-scale clustering were compared for the conditioned and unconditioned cases. Previously we demonstrated (Gerashchenko et al. in J Fluid Mech 668:293–303, 2011; Good et al. in J Fluid Mech 694:371–398, 2012) that in this flow, the overall inertial particle transport is dominated by large-scale intermittent motion corresponding to turbulent bursts penetrating from one side of the mixing interface to the other, and that the particle concentration, to a large extent, preserves its homogeneous turbulence (injection side) values inside of turbulent bursts in the mixing layer. In the present work, we show that the conditioned concentration is higher for the turbulent–non-turbulent than for the turbulent–turbulent interface due to higher averaged burst widths for the latter case. This trend is opposite to that for the unconditioned concentration profiles. The unconditioned particle mass flux peaks approximately in the middle of the layer and is more pronounced for the turbulent–turbulent interface. The conditioned particle mass flux monotonically increases across the layer and is higher for the turbulent–turbulent interface. The small-scale turbulent clustering (less than 10 Kolmogorov scales) quantified by the particle radial distribution function is well preserved inside of bursts. Large-scale clustering (10–500 Kolmogorov scales) caused by the burst activity is observed for the unconditioned cases. Particles with large Stokes number are less sensitive to large-scale clustering than those with small Stokes number.