Litterfall collection is a non-destructive direct method to estimate forest leaf area index (LAI) and validate indirect LAI products. However, the potential errors associated with the variation in specific leaf area (SLA) are rarely explored. Here, we measured the SLA of leaf litter for each tree species in a temperate deciduous forest using the litterfall collection method from 2012 to 2018, and assessed the spatial and temporal variation in SLA and its consequence on LAI estimates. The results showed that the spatial and temporal variation in SLA for the 10 major species across the seven years ranged from 0.8% to 24.3%, with the variation across the nine permanent plots (20 m × 30 m) being higher than that within plot (five 1-m2 traps), and interannual higher than seasonal. Across the 63 plot-years, the 10 simplified SLA sampling protocols introduced the errors in LAI by −12.4% to 22.2% relative to the reference protocol (sampling leaves from each trap at each collection date for the major species). Applying the SLA obtained from one trap in each plot, one plot, at the leaf fall peak for each year, or for a single year induced the errors in LAI by –3.7% to 2.9%, −6.4% to 14.7%, −9.2% to 4.7%, and −8.1% to 8.6%, respectively. Considering the trade-off between inter-plot and interannual variation, we recommend sampling from one-trap for each plot at the leaf-fall peak for measuring SLA and quantifying the spatial pattern of LAI, and sampling each species in the typical plot for measuring SLA and monitoring the temporal fluctuation in LAI. The tight relationship between the MODIS and ground reference LAI across the seven years (R2 = 0.77) validated the use of MODIS LAI to study the long-term change in forest LAI. These findings help to establish a standardized protocol for long-term accurately measuring forest LAI with the litterfall collection.