Software as a man-made system is a typical complex system

understanding its evolution contributes to better software engineering practice. In this paper

we construct software network model from a multi-granularity perspective

namely the level of package

class and feature respectively. Then we analyze the evolutions of three open-source software systems in terms of network scale

quality and structure control indicators

using complex network theory. Finally

taking Lehman's evolution laws as the benchmarks

we compare the evolution of software networks based on multi-granularity. The results show that: (1) the evolution characteristics are varied under different granularity levels

and software network built in the level of class supports the most Lehman laws; (2) the laws of continuing growth

increasing complexity

self-regulation and conservation of familiarity are independent of the levels of granularity; (3) the impact of software evolution in the level of package on software quality is trivial

but feedback system is only supported in the case of class level.