Previous studies have revealed that the predictors of short- and long-term stroke recurrence are different. We designed a comprehensive stroke recurrence (CSR) model, composed of demographic, clinical and radiological findings, to predict long-term ischaemic stroke recurrences. We retrospectively collected the derivation cohort from consecutive patients with first-ever ischaemic stroke within 7days of symptom onset. Univariate and multivariable Cox regression analysis were used to evaluate the association between 2-year recurrence and demographic, clinical and neuroradiological factors. The CSR score was calculated by adding the integer value of independent predictors that was derived from the β-coefficient in the multivariable analysis. To qualify the model, we analyzed the receiver operating characteristics curve. We assessed internal validation with bootstrap methods and assessed external validation with another independent cohort. A total of 958 patients were enrolled, and 63 patients had recurrent strokes during the follow-up periods. The rate of stroke recurrence was 7.0% at 2years. In the multivariable analysis, multiple stage lesions, isolated cortical lesions on diffusion-weighted imaging, severe white matter hyperintensities, multiple lacunar infarctions and relevant arterial stenosis were independently associated with stroke recurrence. The CSR model showed good discrimination [area under the curve (AUC), 0.81 (0.74-0.88)], which was consistent with internal [AUC, 0.75 (0.66-0.85)] and external [AUC, 0.80 (0.69-0.90)] validation. Abnormal neuroimaging findings, rather than cardiovascular risk factors, are predictive of long-term ischaemic stroke recurrence. Causative mechanism of stroke and underlying hostile brain milieu seem to be associated with long-term stroke recurrence.