Elevated metal concentrations in soils can disturb the soil ecosystem; thus, researcherss rive to identify the most sensitive assay for detection of the early signs of toxicity. The purpose of the present study was to compare eight different ecotoxicological endpoints on the same set of metal-contaminated soils that were collected from seven series of soils sampled during field trials. The endpoints are based on three microbial assays (potential nitrification rate [PNR], substrate-induced respiration [SIR], and basal respiration [BR]) and two plant growth tests, one of which included symbiotic N fixation. The overall sensitivity of the endpoints to detect statistically significant adverse effects ranked as follows: PNR > SIR (lag time) > plant yield and N fixation > SIR (respiration after 24 and 48 h) > BR. The lowest adverse effect concentrations were found with the PNR at 7 mg kg−1 of Cd and 107 mg kg−1 of Zn. The variability of these endpoints among different uncontaminated soils was additionally assessed on 14 soil samples. That variability showed a strong correlation with sensitivity scores, illustrating that metal-sensitive endpoints have a large natural variability. We question the ecological relevance of highly sensitive microbial assays, because they tend to have a large natural variability. The identification of toxicity in the field requires endpoints that are highly sensitive and that do not vary greatly among soils (i.e., robust); however, no such endpoint was found in the present study. The endpoints that combined average sensitivity and robustness were SIR (lag time), clover yield, and N fixation in clover.