Language therapy for acquired aphasia has come under much scrutiny in recent years, with debate focusing on whether or not it is effective and how best efficacy can be measured. In this paper, we argue that although the efficacy of many different therapies has been clearly documented, there remain unanswered questions. In particular, it is still difficult to predict which therapeutic task or approach will be successful at remediating which particular disorders. We argue that this is particularly true in the rehabilitation of anomia, whilst in other areas (e.g., reading) the relationship between deficit, task, and outcome is easier to interpret. We discuss in detail the reasons for these differences arguing that specificity of theoretical models, complexity of therapeutic tasks, and a lack of (reporting of) comprehensive assessment of linguistic deficits all contribute to the problem. We conclude that research needs to perform a microanalysis of the interactions between tasks and deficits.