More and more effort is done in BCI research to improve its usability for patients, with respect to its communication speed and transmission accuracy. In this contribution, we ex- periment with BCI speller based on P300 evoked potential. More precisely, the typical form of event-related potential (ERP) inspires us to devise classification methods based on the simi- larity/dissimilarity in the time domain between single trials and one or several estimated ERP templates derived from sub ject recordings. The reliable estimation of template is difficult in a single trial due to the low signal-to-noise ratio (SNR) of electroencephalographic (EEG) signals. We first explicitly estimate the template using several averaging techniques: point- to-point averaging, cross-correlation alignment and dynamic time warping. Then we inexplic- itly estimate several ERP templates using learning vector quantization algorithm combined with an extreme learning machine. Finally classification is realized based on the similar- ity/dissimilarity between the single trials and the template. Simulation is carried out using a BCI competition III data set acquired with the P300 speller paradigm. The experiments show that template-based classifiers can also obtain high accuracy.