For serious accidents such as toxic/flammable release, fire and explosion in chemicals-concentrated areas, one primary issue is to evacuate on-site workers from potential affected zones. In this paper, based on corresponding reports, photos and videos, some accidents involving emergency evacuation are firstly investigated to present important and interesting problems to be considered in practice. These problems are summarized into three kinds of aspects including the poor efficiency of emergency plans, the low accuracy and high time-consuming calculation for evacuation route planning, and the neglection of evacuee's behavior and decision-making. Two main factors influencing the behavior and decision-making of evacuees are concluded as potential risk and impact from hazard sources, and the visibility at different locations. Then, the literature with respect to the emergency evacuation under toxic gas leakage, fire and visibility-limited conditions are reviewed. The key problems are related to ignoring the time-consuming requirement in emergency activities and domino effects of accidents, and the oversimplification of corresponding models. At last, from the famous cellular automata (CA) for emergency evacuation and atmospheric dispersion, a new general research framework for evacuation in chemicals-concentrated areas is proposed. It relies strongly on the quantitative assessment of risk, the establishment of individual visibility field, the use of CA coupled with artificial neural networks, and the optimal route planning. This paper could be useful for realistic problems on emergency evacuation in chemicals-concentrated areas.