Background & aims: Due to the complexity of early diagnostic decision making, we examined the predictive value of an early diagnostic classification and early abilities on later best estimate diagnosis for 22 clinically referred children with language difficulties. Methods and procedures: Four years after initial evaluation (Time 1), the clinical files of these children were reviewed. A best-estimate (BE) diagnosis of language disorder (LD), intellectual disability (ID), or autism spectrum disorder (ASD) was established, with ASD being most common. Outcomes and results: Early clinical classifications were relatively unstable or difficult to establish at a young age. The magnitude of children's cognitive and receptive language delay was a significant predictor of a later BE diagnosis of ID and LD respectively. A BE diagnosis of ASD, by contrast, could not be predicted from children's early social communication problems nor the presence of restricted and repetitive behaviors and interests. Conclusions: Taken together, the results of this study suggest that language difficulties can be an early marker of a neurodevelopmental disorder which is often not identified at the age of first referral. Implications: Eligibility for treatment should, therefore, be based on biopsychosocial case formulation rather than DSM or ICD diagnostic classification. What this paper adds?: In this study a dimensional approach was used to characterize the abilities of young children referred with mild to profound receptive and/or expressive language difficulties. Later on, a categorical approach was adopted to establish best estimate diagnoses. Our clinical, broadly defined sample reflects the heterogeneous intake of young children referred for diagnostic assessment. Other studies on diagnostic stability often only focus on one diagnostic category (and are explicitly excluding children with specific other diagnoses), not taking into account the difficulties of early differential diagnostic decision making and stability across different categories over time. Investigations of differential diagnosis within a clinical group, instead of only differentiating children with a specific diagnosis from typically developing children, may be more informative for clinicians.