Based on efficacy and safety data, several drugs have been approved for symptomatic improvement of dementia of the Alzheimer type and one for the symptomatic improvement of dementia associated with Parkinson's disease. However, established treatment effects must be considered as modest. Randomized clinical trials in other subtypes of dementia (e.g. vascular dementia) have not been able to demonstrate clinically relevant symptomatic improvement, nor has it yet been possible to establish disease-modifying effects in any dementia syndrome or its subtypes. Recent progress in basic science and molecular biology of the dementias has now fostered new interest for more efficacious symptomatic treatments as well as for disease-modifying approaches in the degenerative dementias. For regulatory purposes this requires better standardization and refinement of diagnostic criteria, which allow the study of homogeneous disease populations in specialized academic centers as well as in the general community setting. Depending on the disease stages (early versus late, mild to moderate to severe impairment) and disease entities, distinct assessment tools for cognitive, functional and global endpoints should be used or newly developed. The typical design to show symptomatic improvement is a randomized, double-blind, placebo-controlled, parallel group study comparing change in two primary endpoints, one of them reflecting the cognitive domain and the second preferably reflecting the functional domain of impairment. The changes must be robust and clinically meaningful in favor of active treatment versus placebo. If a treatment claim for prevention of the emergence, slowing or stabilizing deterioration is strived for, it has to be shown that the treatment has an impact on the underlying neurobiology and pathophysiology of the process of dementia. Establishing such an effect in a highly variable progressing syndrome is complex and difficult; however, a variety of trial designs has been provided, including baseline designs, survival designs, randomized start or randomized withdrawal designs, with or without incorporation of biomarkers as surrogate endpoints (e.g. magnetic resonance tomography, emission tomography, cerebrospinal fluid markers). To be accepted as a surrogate endpoint such a biomarker ideally should respond to treatment, predict clinical response and be compellingly related to the pathophysiological process of the dementia. However, careful and sufficient validation of proposed biomarkers as a potential surrogate endpoint is a prerequisite for acceptance by regulatory bodies. This review outlines the regulatory requirements for approval of a new medicinal product for symptomatic improvement or disease-modifying effects in patients with dementia, with special emphasis on the importance of validation of the assessment tools and potential surrogate endpoints based on recent experience and discussion regarding anti-dementia drugs in the European framework.