Observations suggest that the dry-season length (DSL) has increased over southern Amazonia since 1979, primarily owing to a delay of unit ending dates (dry-season end, DSE) and is accompanied by a prolonged fire season. A poleward shift of the subtropical jet over South America and an increase of local convective inhibition energy in austral winter (June-August) seem to cause the delay of the DSE in austral spring (September-November). These changes cannot be simply linked to the variability of the tropical Pacific and Atlantic Oceans. Although they show some resemblance to the effects of anthropogenic forcings reported in the literature.  CMIP5 models substantially underestimate the variability of the DSE and DSL, in part due to models underestimate of poleward shift of the southern hemispheric subtropical jets and increase of convective inhibition energy.  Presumably due to these biases, these models project insignificant change of the DSE and DSL even when these models are forced by RCP8.5 scenario. Our study suggests that the future change of the DSE and DSL may be underestimated by the climate projections provided by the Intergovernmental Panel on Climate Change's Fifth Assessment Report models. Although it is not clear whether the observed increase of the DSL will continue in the future, were it to continue at half the rate of that observed, the long DSL and fire season that contributed to the 2005 drought would become the new norm by the late 21st century. The large uncertainty shown in this study highlights the need for a focused effort to better understand and simulate these changes over southern Amazonia.