The biases for each layer i are net.b{i}. So for a two layer network the biases are net.b{1} and net.b{2}.
The weights to layer i from input j are net.IW{i,j}. For a typical two layer network net.IW{1,1} will exist, while net.IW{2,1} will be empty because the input only goes to layer 1.
The weights to layer i from layer j are net.LW{i,j}. For a typical two layer network net.LW{2,1} will contain the weights to layer 2 from layer 1 and the other layer weights will be empty.
You should also take into account the input and output processing, if you are wanting to reproduce the input-output network function yourself. These functions and settings are available for a two layer network with these properties:
net.inputs{1}.processFcns
net.inputs{1}.processSettings
net.outputs{2}.processFcns
net.outputs{2}.processSettings
