Levodopa is widely used to treat Parkinson’s disease (PD), and its long-term therapy may induce dyskinesia in a dose-dependent manner. However, the threshold dose with a relatively low risk for dyskinesia has not been determined. Demographic, clinical profiles and detailed information of dopaminergic drugs were recorded for 403 PD patients in treatment with levodopa. Variables were compared between dyskinesia and non-dyskinesia groups. Logistic regression analysis was used to assess the association between levodopa dose–related variables and dyskinesia. Receiver operating characteristic curve and decision tree classification model were used to investigate the cut-off value of levodopa dose to best separate the dyskinesia group from the non-dyskinesia group. Patients with dyskinesia tended to have a lower weight and age at onset, higher percentage of female and wearing-off, longer duration of disease and levodopa treatment, higher H-Y stage and MDS-UPDRS Part III score, and higher levodopa dose and levodopa equivalent dose than those without dyskinesia. After adjusted for demographical and clinical variables, levodopa dose–related factors (daily dose, cumulative dose, and weight-adjusted dose) were still associated with dyskinesia. Both the receiver operating characteristic and decision tree classification analysis indicated that patients who have taken levodopa dose ≤ 400 mg per day may be associated with a reduced risk for dyskinesia. In conclusion, we evaluated the thresholds of levodopa treatment with a relatively low risk for dyskinesia. These data should be considered for prevention and management of dyskinesia in patients with PD. © 2019, Fondazione Società Italiana di Neurologia.