diff --git a/NAMESPACE b/NAMESPACE
index 0ce2517..73d973a 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -19,6 +19,7 @@ export(process_tdl_cycle_erml)
 export(read_gm_table)
 export(read_licor_file)
 export(read_tdl_file)
+export(remove_points)
 export(smooth_tdl_data)
 export(specify_variables)
 
diff --git a/R/remove_points.R b/R/remove_points.R
new file mode 100644
index 0000000..ad32c3f
--- /dev/null
+++ b/R/remove_points.R
@@ -0,0 +1,45 @@
+remove_points <- function(exdf_obj, ...) {
+    if (!is.exdf(exdf_obj)) {
+        stop("remove_points requires an exdf object")
+    }
+
+    # Create a list of the optional input arguments
+    arg_list <- list(...)
+
+    # Make sure the optional input arguments are all lists
+    list_check <- sapply(arg_list, function(x) {!is.list(x)})
+    if (any(list_check)) {
+        stop("optional arguments to remove_points must be lists")
+    }
+
+    # Make sure the optional input arguments have names
+    name_check <- sapply(arg_list, function(x) {is.null(names(x))})
+    if (any(name_check)) {
+        stop("optional arguments to remove_points must have names")
+    }
+
+    # Go through each set of conditions to remove the desired points
+    for (point_description in arg_list) {
+        # Make sure the exdf object contains the specified columns
+        required_columns <-
+            lapply(point_description, function(x) {return(NA)})
+
+        check_required_columns(exdf_obj, required_columns)
+
+        # Initialize the logical vector of points to keep
+        points_to_keep <- rep.int(FALSE, nrow(exdf_obj))
+
+        # Apply all the conditions, adding back any points that don't meet the
+        # criteria
+        for (i in seq_along(point_description)) {
+            name <- names(point_description)[i]
+            condition <- point_description[[i]]
+            points_to_keep <- points_to_keep | !(exdf_obj[ , name] %in% condition)
+        }
+
+        # Truncate the exdf to just the points that don't meet the condition
+        exdf_obj <- exdf_obj[points_to_keep, , TRUE]
+    }
+
+    return(exdf_obj)
+}
diff --git a/man/remove_points.Rd b/man/remove_points.Rd
new file mode 100644
index 0000000..66d291c
--- /dev/null
+++ b/man/remove_points.Rd
@@ -0,0 +1,58 @@
+\name{remove_points}
+
+\alias{remove_points}
+
+\title{Remove specific points from an exdf object}
+
+\description{
+  Removes all points from an \code{exdf} object that satisfy a set of
+  conditions.
+}
+
+\usage{
+  remove_points(exdf_obj, \dots)
+}
+
+\arguments{
+  \item{exdf_obj}{An \code{exdf} object.}
+
+  \item{\dots}{
+    Each optional argument should be a list of named elements that specify
+    points to be removed from \code{exdf_obj}. For example,
+    \code{list(species = 'soybean', plot = c('1a', '1b'))} specifies the set of
+    points where (1) \code{species} is \code{'soybean'} and (2) \code{plot} is
+    \code{'1a'} or \code{'1b'}.
+  }
+}
+
+\value{
+  An \code{exdf} object formed from \code{exdf_obj} by removing all rows that
+  meet the conditions specified by the optional input arguments.
+}
+
+\seealso{\code{\link{exdf}}}
+
+\examples{
+# Create an exdf object by reading a Licor Excel file
+licor_file <- read_licor_file(
+  system.file("extdata", "ball_berry_1.xlsx", package = "PhotoGEA"),
+  c(3, 5, 7, 9, 11, 13), 14, 15, 16, 17, 'time'
+)
+
+# Print the number of points in the data set
+nrow(licor_file)
+
+# Remove the following:
+# - All points where `obs` is 28 (1 point)
+# - All points where `species` is `soybean` and `plot` is `1a` or `1b` (14 points)
+licor_file <- remove_points(
+  licor_file,
+  list(obs = 28),
+  list(species = 'soybean', plot = c('1a', '1b'))
+)
+
+# There should now be 15 fewer points remaining in the data set
+nrow(licor_file)
+}
+
+\concept{exdf}
diff --git a/vignettes/analyzing_ball_berry_data.Rmd b/vignettes/analyzing_ball_berry_data.Rmd
index cd357f6..91e11db 100644
--- a/vignettes/analyzing_ball_berry_data.Rmd
+++ b/vignettes/analyzing_ball_berry_data.Rmd
@@ -310,6 +310,46 @@ so it is reasonable to expect that the measurements represent true steady-state
 values. Considering this, all of these curves are acceptable based on the CO~2~
 plots.
 
+### Stability
+
+```{r plot_stability}
+xyplot(
+  `A:OK` + `gsw:OK` + Stable ~ Qin | curve_identifier,
+  data = licor_data$main_data,
+  type = 'b',
+  pch = 16,
+  auto = TRUE,
+  grid = TRUE,
+  xlim = c(0, 2200),
+  xlab = paste0('Incident PPFD (', licor_data$units$Qin, ')')
+)
+```
+
+When measuring response curves with a Licor, it is possible to specify stability
+criteria for each point in addition to minimum and maximum wait times. In other
+words, once the set point for the driving variable is changed, the machine waits
+until the stability criteria are met; there is a minimum waiting period, and
+also a maximum to prevent the machine from waiting for too long. These stability
+criteria are especially important for Ball-Berry curves, since the stomata may
+take a long time to reach steady state.
+
+When these curves were measured, stability criteria were supplied for the net
+assimilation rate `A` and the stomatal conductance `gsw`. The stability status
+for each was stored in the log file because the appropriate logging option for
+stability was set. Now, for each point, it is possible to check whether
+stability was achieved or whether the point was logged because the maximum
+waiting period had been met. If the maximum waiting period is reached and the
+plant has still not stabilized, the data point may be unreliable, so it is very
+important to check this information.
+
+In the plot, `A:OK` indicates whether `A` was stable (0 for no, 1 for yes),
+`gsw:OK` indicates whether `gsw` was stable (0 for no, 1 for yes), and
+`Stable` indicates the total number of stability conditions that were met. So,
+we are looking for points where `Stable` is 2. Otherwise, we can check the other
+traces to see whether `A` or `gsw` was unstable. Here it looks like many of the
+high light points were not stable, and it may be a good idea to remove them
+before proceeding with the Ball-Berry fitting.
+
 ### Light-Response Curves
 
 ```{r plot_assimilation}
@@ -340,10 +380,31 @@ likely a byproduct of the noise that was intentionally added to the true
 measured data (see [The Data]). Nevertheless, it may be a good idea to remove
 them before proceeding with the Ball-Berry fitting.
 
-## Removing Bad Curves
+## Cleaning the Licor Data
+
+While checking over the plots in [Qualitative Checks], two issues were noticed:
+(1) some points were logged before stability was achieved and (2) some of the
+curves look abnormal. In this section, we will demonstrate how to remove the
+unstable points and the weird curves.
+
+The following command will keep only the points where `Stable` is exactly 2;
+this condition means that all of the stability criteria were satisfied.
+Sometimes, following this procedure, a curve will have very few stable points
+remaining; it is usually a good idea to automatically exclude any curve with
+fewer than three stable points.
+
+```{r remove_unstable_points}
+# Only keep points where stability was achieved
+licor_data <- licor_data[licor_data[, 'Stable'] == 2, , TRUE]
+
+# Remove any curves that have fewer than three remaining points
+npts <- by(licor_data, licor_data[, 'curve_identifier'], nrow)
+ids_to_keep <- names(npts[npts > 2])
+licor_data <- licor_data[licor_data[, 'curve_identifier'] %in% ids_to_keep, , TRUE]
+```
 
 Since we have identified a few curves that may not be acceptable for Ball-Berry
-fitting, we can remove them as follows:
+fitting, we can remove them via the `remove_points` function from `PhotoGEA`:
 
 ```{r remove_curves}
 # Define a list of curves to remove from the data set
@@ -355,9 +416,13 @@ curves_to_remove <- c(
 
 # Remove them
 licor_data <-
-  licor_data[!licor_data[, 'curve_identifier'] %in% curves_to_remove, , TRUE]
+  remove_points(licor_data, list(curve_identifier = curves_to_remove))
 ```
 
+Note that `remove_points` can also be used to exclude individual points instead
+of entire curves; see its help entry by typing `?remove_points` for more
+information.
+
 # Fitting Licor Data
 
 Now that we have checked the data quality, we are ready to perform the fitting.
@@ -554,6 +619,8 @@ file to initialize your own script.
 
 <<plot_assimilation>>
 
+<<remove_unstable_points>>
+
 <<remove_curves>>
 
 <<calculate_bb_index>>
diff --git a/vignettes/analyzing_tdl_data.Rmd b/vignettes/analyzing_tdl_data.Rmd
index 6704d00..7f3ecf4 100644
--- a/vignettes/analyzing_tdl_data.Rmd
+++ b/vignettes/analyzing_tdl_data.Rmd
@@ -349,9 +349,10 @@ system, and there is evidence that the calibration may be compromised during the
 spike, it is probably a good idea to remove the TDL cycles where the ^12^C
 signal from valve 26 suddenly becomes large.
 
-Fortunately, this is simple to do in R. The following illustrates one way to
-"clean" the data by removing the unreliable TDL cycles; here, we actually remove
-two regions of the data that appear to be suspicious:
+Fortunately, this is simple to do using the `remove_points` function from
+`PhotoGEA`. The following illustrates how to "clean" the data by removing the
+unreliable TDL cycles; here, we actually remove two regions of the data that
+appear to be suspicious:
 
 ```{r removing_cycles}
 # Define a vector of cycle numbers that should be removed
@@ -362,7 +363,7 @@ tdl_cycles_to_remove <- c(
 
 # Remove them
 tdl_files_clean <-
-  tdl_files[!tdl_files[, 'cycle_num'] %in% tdl_cycles_to_remove, , TRUE]
+  remove_points(tdl_files, list(cycle_num = tdl_cycles_to_remove))
 ```
 
 ## Smoothing the TDL Data