Sufficient soil phosphorus (P) content is essential for achieving optimal crop yields, but accumulation of P in the soil due to excessive P applications can cause a risk of P loss and contribute to eutrophication of surface waters. Determination of a critical soil P value is fundamental for making appropriate P fertilization recommendations to ensure safety of both environment and crop production. In this study, agronomic and environmental critical P levels were determined by using linear-linear and linear-plateau models, and two segment linear model, for a maize (Zea mays L.)-winter wheat (Triticum aestivum L.) rotation system based on a 22-yr field experiment on a Haplic Luvisol soil in northern China. This study included six treatments: control (unfertilized), no P (NoP), application of mineral P fertilizer (MinP), MinP plus return of maize straw (MinP+StrP), MinP plus low rate of farmyard swine manure (MinP+L.Man) and MinP plus high rate of manure (MinP+ H.Man). Based on the two models, the mean agronomic critical levels of soil Olsen-P for optimal maize and wheat yields were 12.3 and 12.8 mg kg−1, respectively. The environmental critical P value as an indicator for P leaching was 30.6 mg Olsen-P kg−1, which was 2.4 times higher than the agronomic critical P value (on average 12.5 mg P kg−1). It was calculated that soil Olsen-P content would reach the environmental critical P value in 41 years in the MinP treatment, but in only 5–6 years in the two manure treatments. Application of manure could significantly raise soil Olsen-P content and cause an obvious risk of P leaching. In conclusion, the threshold range of soil Olsen-P is from 12.5 to 30.6 mg P kg−1 to optimize crop yields and meanwhile maintain relatively low risk of P leaching in Haplic Luvisol soil, northern China.