SummaryThis paper proposes an ideal rateless codes model to comprehensively describe rateless codes and extends the definition of systematic linear block code to generalized systematic code. Under the proposed model, average delay, maximum disorder, and uniformity recovery entropy are introduced as performance indices to design efficient rateless codes. A novel coding scheme based on two‐stage encoding and forward equal probability is proposed to optimize the proposed indices. In the first stage, the first k symbols are encoded, aiming to improve the performance of order recovery, uniformity recovery, and transmission efficiency as much as possible. In the second stage, the remaining infinite symbols are encoded, where the symbols with high degree are used to compensate the symbols loss in the first stage. Simulation results show that the proposed scheme can achieve generalized systematic rateless codes with high probability and also has less average delay and maximum disorder, better capacity achievability, and uniformity recovery performance than Luby trasform (LT) codes and rateless coded symbol sorting algorithm. Besides, the proposed scheme has the aforementioned advantages compared with expending window fountain codes except for uniformity recovery and maximum disorder performance when the erasure rate is higher than about 0.25. Copyright © 2013 John Wiley & Sons, Ltd.