Pseudaletia unipuncta (Haworth) is a pest in Azorean pastures that causes up to 8% yield loss mainly during summer and early autumn. The objective of this research was to develop robust sampling plans based on larval spatial distribution. The number of larvae in three 60-m² plots on S. Miguel Island (Azores) was determined. Plots of 0.25 m² were the most accurate unit for estimating the population size. The relationship between mean and variance fitted both Taylor’s power law and Iwao’s patchiness regression model. Relationships to determine optimum sample sizes for fixed levels of precision based on both models were developed, but demanded a heavy sampling cost for the usual precision values (0.1–0.25). To reduce sampling effort, two sequential sampling plans were developed and compared, one based on Taylor’s parameters and the other based on Iwao’s parameters. For a precision of 0.25, Taylor’s sequential sampling plan led to an average 76% reduction of the sampling effort compared with sample sizes estimated for fixed levels of precision. Simulations of Iwao’s sequential sampling plan applied to larval counts correctly predicted a “treat” or “not-treat” decision for 91% of the cases. However, this plan estimated field densities with a lower degree of precision than Taylor’s plan and required a considerable increase in sampling effort for larval densities close to the critical mean. Use of Taylor’s sequential sampling plan should provide effective management of P. unipuncta in grass pastures and minimize sampling time and cost.