Aims. We study how IMF sampling affects the ionizing flux and emission line spectra of low mass stellar clusters. Methods. We performed 2 × 10 6 Monte Carlo simulations of zero-age solar-metallicity stellar clusters covering the 20−10 6 M mass range. We study the distribution of cluster stellar masses, M clus , ionizing fluxes, Q(H 0 ), and effective temperatures, T clus eff . We compute photoionization models that broadly describe the results of the simulations and compare them with photoionization grids.Results. Our main results are: (a) a large number of low mass clusters (80% for M clus = 100 M ) are unable to form an H ii region. (b) There are a few overluminous stellar clusters that form H ii regions. These overluminous clusters preserve statistically the mean value of Q(H 0 ) obtained by synthesis models, but the mean value cannot be used as a description of particular clusters. (c) The ionizing continuum of clusters with M clus < ∼ 10 4 M is more accurately described by an individual star with self-consistent effective temperature (T * eff ) and Q(H 0 ) than by the ensemble of stars (or a cluster T clus eff ) produced by synthesis models. (d) Photoionization grids of stellar clusters cannot be used to derive the global properties of low mass clusters. Conclusions. Although variations in the upper mass limit, m up , of the IMF would reproduce the effects of IMF sampling, we find that an ad hoc law that relates m up to M clus in the modeling of stellar clusters is useless, since: (a) it does not cover the whole range of possible cases; and (b) the modeling of stellar clusters with an IMF is motivated by the need to derive the global properties of the cluster: however, in clusters affected by sampling effects we have no access to global information of the cluster but only particular information about a few individual stars.