[steven114]

From the chip docs:

PIC18F6520/8520/6620/8620/6720/8720
3.0 RESET
The PIC18FXX20 devices differentiate between
various kinds of Reset:
a) Power-on Reset (POR)
b) MCLR Reset during normal operation
c) MCLR Reset during Sleep
d) Watchdog Timer (WDT) Reset (during normal
operation)
e) Programmable Brown-out Reset (PBOR)
f) RESET Instruction
g) Stack Full Reset
h) Stack Underflow Reset
Most registers are unaffected by a Reset. Their status
is unknown on POR and unchanged by all other
Resets. The other registers are forced to a “Reset
state” on Power-on Reset, MCLR, WDT Reset, Brownout
Reset, MCLR Reset during Sleep and by the
RESET instruction.
Most registers are not affected by a WDT wake-up,
since this is viewed as the resumption of normal
operation. Status bits from the RCON register, RI, TO,
PD, POR and BOR, are set or cleared differently in
different Reset situations, as indicated in Table 3-2.
These bits are used in software to determine the nature
of the Reset. See Table 3-3 for a full description of the
Reset states of all registers.
A simplified block diagram of the On-Chip Reset Circuit
is shown in Figure 3-1.
The Enhanced MCU devices have a MCLR noise filter
in the MCLR Reset path. The filter will detect and
ignore small pulses. The MCLR pin is not driven low by
any internal Resets, including the WDT.

Looks like the most likely reasons (assuming of course that it is actually a reset, which it might not be) would be the stack over/underflow resets. Are you doing anything funny in assembly?