A sequential decision plan was developed for controlling larvae of the bertha armyworm, Mamestra configurata Walker (Lepidoptera: Noctuidae), in canola (Brassica napus L. and B. rapa L., Brassicaceae), using 0.25 and 0.5 m² sampling units. Fields in Manitoba were sampled from 1980 to 1994 to determine minimum sample sizes and upper and lower cumulative larval counts at three economic thresholds. Taylor's power law described most of the variation between mean larval density and variance for 0.25 m² (r² = 0.926) and 0.5 m² (r² = 0.924) samples. Larvae were found to have a moderately clumped distribution in canola (b = 1.42). Levels of precision (D₀) varying from 0.15 to 0.25 caused minimum sample sizes to vary between 6 and 21 for the 0.5 m² samples to between 9 and 31 for the 0.25 m² samples, for an economic threshold of 16–24 larvae/m² (P = 0.20). Mean sampling times ranged from 40–108 for the 0.25 m² samples to 49–126 min for the 0.5 m² samples. The sampling plan for the 0.25 m² samples was verified in 18 fields in 2006 and 2007. A correct decision was made in 87% (D₀ = 0.25), 91% (A) = 0.20), and 94% (D₀ = 0.15) of the fields when the recommendation was to spray if a decision could not be reached after a second sampling. The mean number of samples needed for making a decision was 14 (D₀ = 0.25), 19 (D₀ = 0.20), and 32 (D₀ = 0.15). We recommend that growers use a precision level of 0.20 to minimize error rates and sampling effort. In most years, the minimum number of 0.25 m² samples per field that growers would need to take is 14–17.