Abstract : This thesis proposes and evaluates some online algorithms for machine scheduling problems. Deterministic scheduling models have been extensively studied in the literature. One of the basic assumptions of these models is that all the information is known in advance. However, this assumption is usually not realistic. This observation promotes the emergence of online scheduling. In online scheduling problems, an online algorithm has to make decisions without future information. Competitive analysis is a method invented for analyzing online algorithms, in which the performance of an online algorithm (which must satisfy an unpredictable sequence of requests, completing each request without being able to see the future) is compared with the performance of an a posteriori optimal solution where the sequence of requests is known. In the framework of competitive analysis, the performance of an online algorithm is measured by its competitive ratio. We mainly deal with two online paradigms: the one where jobs arrive over list and the one where jobs arrive over time. Based on these two paradigms, we consider different models: single machine, two identical parallel machines, two uniform parallel machines, batch processing machine and open shop. For each of the problems, we prove a lower bound of competitive ratios and propose online algorithms. Then we further measure the worst case performance of these algorithms. For some problems, we can show that the algorithms we proposed are optimal in the sense that their competitive ratios match the lower bounds.