The executive function (EF) theory of autism has received much support recently from a growing number of studies. However, executive impairments have not always been easy to identify consistently and so novel "ecologically valid" tests have been designed which tap into real-life scenarios that are relevant to and representative of everyday behavior. One characteristic of many of these tasks is that they present the participant with an "ill-structured" or "open-ended" situation. Here, we investigated the possibility that tasks with greater degrees of open-endedness might prove more sensitive to detecting executive impairment in autism. Forty-five children with autism spectrum disorder (ASD) were compared to 27 age- and IQ-matched control children on a range of cognitive tests of EF. Group differences were found on half of the tasks, with the greatest degree of impairment detected on the more open-ended tasks. The ASD group also performed more poorly on a simple control condition of a task. Detailed consideration of task performance suggested that the ASD group tended to create fewer spontaneous strategies and exhibit more idiosyncratic behavior, which particularly disadvantaged them on the more open-ended tasks. These kinds of behaviors have been reported in studies of neurological patients with frontal lobe involvement, prima facie suggesting a link between the scientific fields. However, we suggest that this behavior might equally result from a poor understanding of the implicit demands made by the experimenter in open-ended test situations, due to the socio-communicative difficulties of these children.