Hello, I would like to calibrate a term structure model e.g. Vasicek to market data. I see a lot of people use time series data and do a parameter estimation.But what is wrong with just taking a cross section of option prices and minimising square errors using fminsearch in Matlab?Would anybody know please?

Whether you calibrate to time series data or option/swaption prices depends on what you are using the model for. Calibrate to options for pricing, calibrate to time series for trading.

Thanks, that makes sense. I was confused by the idea of miss-specified models and having multiple parameter selections that could give you the same squared-error. I thought maybe that parameter estimation using a time-series (as apposed to a calibration to a cross section) solved this problem. I was just reading Cont-Tankov though and chapter 13 has put some of my fears to rest.I think minimising squared errors is OK (obviously your parameters will be risk-neutral and not suitable for trading) but you need to be careful about local minimums. This is something I have to think about then.