Abstract: A good many legacy applications are widely used in various scientific research fields.In pre-gridservice times, conventional means of sharing scientific computing resources is by sharing programs and data.With the introduction and wide adoption of OGSA framework and OGSI specification, grid computing has entered a new era of service-orientation, in which implementation of sharing domain expertise resorts to portable, interoperable grid services.So comes the challenge to gridify （to convert to OGSI-compliant grid services） existing scientific computing resources, which is mainly made up of legacy applications.Focusing on this topic, features of scientific legacy applications and key issues during gridification are analyzed, and resolutions are proposed, including a common abstract invocation interface for various applications, an "on-demand" factory responsible for creating and publishing scientific computing services, and a uniform monitor and manipulation interface.The design scheme consists of three parts: scientific computing services which implement three interfaces of GridService, LegacyCodePortType and ManipulationPortType; scientific computing operation providers（OPs）; “on-demand” scientific computing factories.During the course of implementation, a set of difficulties are resolved, including lower-level wrapping and non-blocking invocation.The design solution has been successfully applied to the development work of ChinaGrid project.Application scope tests show that the design solution can apply to most types of scientific computing legacy applications.Performance experiments show that the proposed gridification mechanisms bring about only slight efficiency loss.