This PhD thesis presents a suite of methods for optimising design and for analysing blocking probabilities of all-optical networks. It thus contributes methodical knowledge to the field of computer assisted planning of optical networks. A two-stage greenfield optical network design optimiser is developed, based on shortest-path algorithms and a comparatively new metaheuristic called simulated allocation. It is able to handle design of all-optical mesh networks with optical cross-connects, considers duct as well as fibre and node costs, and can also design protected networks. The method is assessed through various experiments and is shown to produce good results and to be able to scale up to networks of realistic sizes. A novel method, subpath wavelength grouping, for routing connections in a multigranular all-optical network where several wavelengths can be grouped and switched at band and fibre level is presented. The method uses an unorthodox routing strategy focusing on common subpaths rather than individual connections, and strives to minimise switch port count as well as fibre usage. It is shown to produce cheaper network designs than previous methods when fibre costs are comparatively high. A new optical network concept, the synchronous optical hierarchy, is proposed, in which wavelengths are subdivided into timeslots to match the traffic granularity. Various theoretical properties of this concept are investigated and compared in simulation studies. An integer linear programming model for optical ring network design is presented. Manually designed real world ring networks are studied and it is found that the model can lead to cheaper network design. Moreover, ring and mesh network architectures are compared using real world costs, and it is found that optical cross-connects should be drastically cheaper if they are to compete on the cost with ring networks. An MPXS (multi-protocol wavelength switching) simulator is constructed, focusing especially on the timing in the setup phase to asess the blocking probability effect of node and link delays as well as wavelength reservation strategies. Initial results suggest that node delays have a greater effect than link delays, and that only few wavelengths should be reserved during path setup. Finally, a novel way of calculating blocking probabilities in all-opticalWDM(wavelength division multiplex) networks by re-using existing blocking probability calculation tools is presented. The WDM network problem is transformed into a graph colouring problem whereupon information required by existing tools is extracted.