The responses you see in the ERP are usually from the generation of a dipole, the field of which is picked up at different electrode locations.  Thus, some will pick up the positive end, and others will pick up the negative end.  

The shape of the response that you get depends on many factors and, among them, on the reference that you use and on the baseline correction. 

As general rule the acquisition reference should be on extra-cefalic (it is suggested on the nose and it suits more or less all the erp). Then if you want to apply source reconstruction, it is advisable to move to average reference.

Regarding baseline correction there is not a clear rule. The crucial point is to design a task. Indeed, the inter stimulus interval should be long enough to guarantee a period during which there is no EEG change time locked to the event: that can be the baseline that you use. A jitter is advisable as well. Hope it helps

Dear Giovanni Pellegrin,

Yes, I rereference and used average reference. also regarding ISI it was 1500 msec, wasn't that enough?

Usually it's enough but it depends on the ERP...there are some which last longer...however it depends on what you are interested in. 

Dear Nik, can you give more explanation please?

Two interrelated things:

(1) If you're using an average reference, then of course any time that half of your electrodes are showing a positive response, the other half must show a negative response, in order for the average to remain the same.

(2) What Nik said about dipoles. EEG is picking up the electrical field generated by neurons, right, and that electrical field has a negative end and a positive end, which are directed 180 degrees away from each other. If the dipole is oriented so that you're picking up positive voltage at the posterior of the skull, then you're likely picking up negative voltage at the anterior end. 

Without looking at what you're looking at, my money's on it being the "average reference" effect.

Dear Abigail Noyce , thanks for you comment

Any reference paper please?

Books, not papers, but I would recommend both:

Nunez & Srinivasan, Electric Fields of the Brain

and

Luck, Introduction to the ERP Technique.

Purves et al, Principles of Cognitive Neuroscience.  Another good text

Carlson , Physiology of Behaviour.  This might help as well.

This is often called phase reversal as the phase of your waveform is 180 degrees reversed. As mentioned above, if you have an electrical generator (dipole) in the brain, you will see a phase reversal at oposing sites of your surface head.

Dear Stephan, thanks for your comment

Yes, it's clear but I meant why the anterior part responses is a mirror (same exactly with reversed phase) responses of posterior part in all electrodes?

Due to volume conduction you will "see" the effects in many electrodes. Best, Stephan

Positive and negative signal depends on which way the dipole producing the signal is pointed in the brain.  For instance if a population of neurons are producing a positive signal in one area of the brain there will be a negative signal 180 degrees in the opposite direction...just like any electrical circuit.  For instance you can see a positive signal in occipital electrodes and 180 degrees opposite this signal you will see a negative signal in frontal electrodes.  As Stephen mentioned this signals travels via volume conduction.  This information is covered in detail in Steve Luck's ERP book http://mitpress.mit.edu/books/introduction-event-related-potential-technique

Search for examples and figures about "ERP mapping". That will help to understand Mr Morrison's explanation. Good luck!

 Question for Abigail: If average-referencing means that opposite ERP patterns are observed for sites that are on opposite regions of the scalp, does this not mean we get significant effects in these regions as a consequence? For example, if I observe a LAN (negativity) in central-left sites, does this mean I'll see a significant positive deflections in right-central sites (solely as a consequence of negativity at left-central sites)? Does this throw up problems for analysis?

Look up the polar average referencing effect.  The problem won't entirely go away if you use more electrodes, but it will help (like 64+).

If this has/is causing major problems then you can mention the components that you are interested in, as well as this referencing/dipole effect and just measure them independently of their sign (use the rectified /absolute potential).

Can I pose an example? I got a signal from cetro-parietal electrodes that revealed prominent P2, N2 and P300 (both P3a and P3b) peaks.

However, all the polarities were reversed. Can I assume that it does not matter and just flip the y axis?