Schizophrenia is characterized by three major symptom classes: positive symptoms, negative symptoms, and cognitive deficits. Classical antipsychotics (phenothiazines, thioxanthenes, and butyrophenones) are effective against positive symptoms but induce major side effects, in particular, extrapyramidal symptoms (EPS). The discovery of clozapine, which does not induce EPS and is thought effective against all three classes of symptom, has driven research for novel antipsychotics with a wider activity spectrum and lower EPS liability. To increase predictiveness, current efforts aim to develop translational models where direct parallels can be drawn between the processes studied in animals and in humans. The present article reviews existing procedures in animals for their ability to predict compound efficacy and EPS liability in relation to their translational validity. Rodent models of positive symptoms include procedures related to dysfunction in central dopamine and glutamatergic (N-methyl-D-aspartate) and serotonin (5-hydroxytryptamine) neurotransmission. Procedures for evaluating negative symptoms include rodent models of anhedonia, affective flattening, and diminished social interaction. Cognitive deficits can be assessed in rodent models of attention (prepulse inhibition) and of learning/memory (object and social recognition, Morris water maze and operant-delayed alternation). The relevance of the conditioned avoidance response is also discussed. A final section reviews procedures for assessing EPS liability, in particular, parkinsonism (catalepsy in rodents), acute dystonia (purposeless chewing in rodents, dystonia in monkeys), akathisia (defecation in rodents), and tardive dyskinesia (long-term antipsychotic treatment in rodents and monkeys). It is concluded that, with notable exceptions (attention, learning/memory, EPS liability), current predictive models for antipsychotics fall short of clear translational validity.