Abstract Precise yet time-efficient sampling methods for coccinellids would be useful in integrated pest management (IPM) research and decision-making in alfalfa. We compared quadrat sampling, removal sampling, counting coccinellids for a constant amount of time while walking slowly through a field (timed counts), and sweepnet sampling for estimating adult and larval coccinellid density in alfalfa. Removal sampling gave biased estimates of density of adultColeomegilla maculataDegeer, but gave acceptable estimates of density of adults of other species. Regression models were developed to convert relative population estimates of adult and larval coccinellids obtained by sweepnet and timed count sampling to absolute population estimates (number per m2). For adults, plant height was included in the best regression models for converting sweepnet catch and timed counts to population density. For larvae, plant growth stage was included in the best regression model for converting sweepnet catch to density. The number of aphids per 10 sweeps was included in the model for converting timed counts of larvae to density; however, this model was of little value because it explained only 13% of the variation in the relationship between timed counts and population density. Coefficients of determination of regression models for estimating adult density from timed count and sweepnet sampling were 0.90 and 0.93, respectively. The model for converting sweepnet catch of larvae had a coefficient of determination of 0.71. In terms of statistical precision achieved per unit time spent sampling, a 25-sweep sample was the most efficient sample unit for estimating adult density. A double 12-min removal sample was most efficient for sampling larvae, but this method gave biased estimates of larval density, and overall, sweepnet sampling was better.