Forest management practices may alter soil properties, but experimental evaluation of treatment effects and observation of temporal changes are often difficult, because substantial soil variability yields poor statistical sensitivity. This study examined our ability to detect experimental treatment differences in three contrasting Pacific Northwest 60-100 year-old coniferous forests at the Olympic Peninsula of Washington, Cascade Mountains of Oregon, and Siskiyou Mountains of Oregon. The experimental treatments are an unmanaged control and six manipulations of overstory species and woody debris on 6-ha plots, replicated in three or four blocks at each forest. If only post-treatment measurements had been made, the minimum detectable differences between extreme treatments would be 20 to 60% of initial soil C and N. If pretreatment measurements were used as covariates, minimum detectable differences for C would remain the same, but those for N would decrease by as much as half, because N has greater pretreatment variability among plots than C.